Add to Watchlist
Multiresolution Analysis, Form based Features, Thresholding, Edge Detection, Morphological Operators (28.04.2011)
Series
Annotations
Transcript
00:00
So what I'm all to on new lecture multimedia databases and the today want to talk to you about a couple of things last time and we rode the with textured so we were talking about what makes attacked what makes a structure of both of them some some some image and were considering some from load of the feature on 1 end like like contrast all direct melody and we will be talking about some probabilistic ways of dealing with with Texas so we could predicts the colour of the intensity of the some pixel based on their surroundings of that pixelated the pattern and we were talking about high level features about Fourier transformation about a leaflet transfer owns bomb to give you an insight into what we can gain from other spaces like features spaces of frequency spaces out what evident beat that the transport transport all in which the difference between high level foetuses load of the features
01:15
The major difference anybody But During Nobody Are basically from high level features without losing any of the image information you can reconstruct the texture As just a different kind of and for Oliver if you use a the future your extracting some characteristics some measurements on the structure and the usually impossible to reconstruct real regional image the origin of structure from low level feature of the basic difference between 2 days we want to go a little bit into into into shapes we want to see I'm FVB highly can put all fingers on an object that opera in images and the shape of a best description of a description of what was actually is in the image and them that we will work a little bit on offer much resolution of the House and then go to the real shape a features of was of different ways of preparing off the commanding the image and detecting the basic at the city and at just that may make actually out the visual impression of the of the show
02:45
But 1st we won't stop eluded about my 2 resolution and the and the and the problem that we see the last time was that even full of the smell that functions that we useful follow offload image frequency for the intensity and the Distribution found the transformations can become are the obdurately complex wrapped up and we see the with the transfer on which is debt 7 different different ways that will used among wavelet 2 children and and 4 debut with the and even that was a high demand for the patient systems and and if we have not sell a bigger base for all features that this is getting even more complex and we have a lot of information the image on 1 hand we have a lot of of variables given by the face vector or given by the different frequencies in the assignment around the world waves and a slow in this Linear equation is definitely a very expensive think so that what we do with all we or any did go into some algorithms like The The Fastest Fourier Transform that makes it a little bit easier to calculate the actual Linear Systems
04:27
With the recent and then we have something similar which is called The Last which from who would of notice In and out on to show you a Hall this works to because the of this is basically a on the road of the foundation of multi resolution and and that is that you want after topics 0 1 1 classical ways of dealing with high demand computations that he will from the experience of very off name in the UK The and and and before the Bill was passed with the terms for a new can actually compute the result of Philip Linear patients system Linear time and only basically do if you change between 2 steps you Convert the image found into reduced resolution to get smaller images was last resolution and you start a 2nd thought and that is what you lost during the reduced their the reductions the icons Represent images the ways by 2 possible the basic features And the details And this is where we get in the party of the are so called the basic features are so visible in below and Load a resident of reserves of representation are used to see the big edges is to see the dominant colours But if you draw a deeper into the image you find that this is not actually thorough of the same colour But the cover changes slightly and these changes Could be Stalled in a different I'm part of the image or different from representation of the game and was these 2 images you connect the reconstruct the original image By the adding the details To the stadium pop up there And this is basically but was called mighty resolution and losers so you have a picture in many resolution and for each resolution That is smaller than the origin resolution All the details that could be used to reconstruct fioritura resolution 2 but in a different So you get 1 found that just shows that the basic get another file that he can for reconstruction if you want to Popular
07:39
This particular itself where you do is you consider the image in different resolution and the stigma of being a justice the intensity signalled its you go through the image and your a called the intensity of each left I do well you can Start with a single picks A single Pixo has just a single intensity The violent at more details have to say what actually It was a single Pixo it will 6 picks and the Distribution off the brightness all colours in this week so I was like that end something you Which goes to want the construction If I'd do that several times to ironed out at the original image up and this is the basic ideas so the representation of the image in certain blokes in certain resolution It is just time By averaging The intensity information And keep the difference is what he used for every may be an example will make it more clear
09:16
What you basically do what they do You taking The big picture That might 16 Picks up In the 1st step will always take the pixels Every which they intensity value and stole them as a single pigs Oka It led to a couple of times with another 1 pigs Press The something I've to nobody is of what happened here What do loos by the average told the origin of the new look like Kent and his commit just under by differences surpassed on it
10:17
They've rested image Became and said Became minus 1 is just low resolution where the fixes have been than the average and at some point we arrive at the great though the worst revolution resolution about just a single pigs up And the single Pixar Pixo shows me the average colour Note that in round a when we were talking about about the foetus for colour Everage colour was a necessary and a meaningful which in a way so even This image tells us something About the it gives us a way to distinguish between image because if the the images were totally different to stop with the also there are also that that the average colours are different but she's not is this this way around is not as true as the other way around If the average colour of to images different than the origin of images must have been before It was that at different can still come down to the same average So we have the examples of and a mixed being being half red and how blue and an image being totally lilac by their the same ever sculpted but still that different images that if the UK which would be green no way of getting the same as basically the and back But still we have to go back from The singer image tool reconstructed so we still have to ask about what's happening he had we go back and that is basically the intensity of pixels off a obtained from the coast How does the back transformation work and family
12:50
Basically The intensity of the peaks in the low grade the means of a set of corresponding pixels in the high resolution great Foca and Whisker 1 step by high the number of pixels So Apple with half the number of pixels to pixels become what they and if to pixels become 1 When storing the inflammation Contained in the 2 Pixo can basically be done Howell But the IAAF to pixels intensity I'm intensity Jake and a goal to 1 Pixo which as intensity Yes There I've just days divided by 2 basically a pin just the average What do after struggle is our so 1 to have this step But It Yes on the Exactly the difference between night and day because they find their what the mood and the and the difference was restricted of the Put my fingers on what they she were if the difference 0pc there must have been exactly the same colour on the spot if the difference to the mass of been ride 1 low on higher folk and so this is actually want to do not I'm Starring the difference He UK and and this would make so that happens basically is ideal from 200 300 200 to 150 200 from 150 to to run to a 100 for the year so I'm was kind of like the boiling it down in
15:32
The party is under way and in the word of a pet this allways were to cope from Saudi 100 can I read and 100 200 simple down to 150 200 size template downed and the vertical accents and and from 150 200 assembly down to of 150 100 up so now it exactly equipped of the size of the original image of this basically is hot vertically and half horizontal and this is bygone and as you can see it is getting closer and cost of the not see out the The the actual the actual objects and more at some point but you just see very close representation of the future now for each Pixo in the offer to be made
16:39
There was a corresponding pixel In below resolution rusted this to write to By repeated averaging and If I'm the intent city of a pixel at some point and for each takes we have what is left of the rich part class the different that we use to kind of like simply don't and and this is the tale information that is needed to reconstruct the intensity of the actual picture of the actual picture and if where case that the way in resolution in terms of resolution from the Oriental pictured with left to educate times the differences and then we find the arrived at the origin of picked So this is last less decomposition though if we only look at that The 1st part it very lousy We can do last things with the further cost which is interesting but we need the detailed information to kind of like time so and 1 of the ways that was basically already hinted at averaging and different 6 or what we do but we take each to fix also entrance from them into the Duke of Kent
18:16
Easy Then we take the next 2 Pixar Owais horizontally and the and The ship Take them to a mix of but what do we have to do basically in if I'd duty on the fight and the mind To have an average of 7 And after saying how with the different from each other they give them again Well basically the 1st 1 of the 2nd 1 is 7 last to buy keep the to 4 9 7 plus 2 If I'd do 11 9 and want to arrive at an average of 7 what will be other number be Of 55 of the idea the same for the aid and the fit again for the 2nd number by recalled the different with respect to the average which this well and 6 makes minus to talk and reclad the mind to what his eye have to think what we were was the the 1st number Well have the 6 as an average of the flow as 1 of the pubs and the other part has to be a again I've to decide for even want it works either way just take want you can stand the difference was respectively 1st ball was offered to 2nd doesn't matter but it has to be the same way as obviously The 5 foot benign and now because what basically it you were waste put together 2 pixels to fail 1 average picks up here And the information that you need This depending on the average obviously And on the difference was respected either of the number numbers it doesn't matter if you choose the 1st number of number talking playground and stay with a study confusing if you Dhirendra need BBC any number Unite Benidorm work that occurred immediately say well defined 7 this the average and the and the night is 1 of the numbers and and the and the other members of the have to do this solution Which is easy But Basically what a candle but but you have to decide at 1 point but I don't want the demand which take but for you want and you can see now is that we keep the every just And we keep The differences A camp It's the same amount of information It is now So what we saving Numbers you aged numbers to you Copes act And 8 numbers here For And replaced 16 numbers by 2 times the number yet What do we say it welcomed the group yes to you do get a preview of the if you just skip the details and the and use this yes we can do something with a silly of to stall same out of date and only Tricky P Yes there will be a while P The moral of the tale is that it Yet that was the basically ideological a high demand of a solo we can we can see things and in the end of a long resolution version of that I'd interesting phosphorite and this actually doing but still as that they have arguement about the storage Yes but yes the exactly if we start The differences with respect to the average its from old numbers you and most of them will actually be Zero because they got pixels and in some image you relentless if you guys have the Orange Fletcher now like these big so that will do well in in the same colour differences hero will be very off in the same difference and available was not right and was indifferent and happens 1 black picks up the difference This is the interest in part of this is where the actual compression comes from many of these are 0 But and this is kind of like the but there with storey 1 at some point you arrive the single Pixo P update said on Friday the average intensity of this image of it looks like for or Kent Everybody understood why I'm doing Get
24:52
So what I'm doing In signalprocessing that has actually something that is the even even nicer to do then the averaging and and and simply what you could do is to say I'm hypothesise load of the year a signal that this year we I you and what you do with with a high cost and low cost of as the hype after the with only egullet high frequencies low part of with only figure The load Africa it for this basic the what happens with the help of those would that of the man and is that what happened was the hype part of the and and the hypothesis allways basically out the baby with the of high all of them And the low cost this the baby wave of law that the low cost information contains a says the and inflammation and the average us the high possibly fulmination contains the detailing fulmination the difference care But
26:12
I'd have been expecting which details low cost of the average end of cost you could use possible Applications off the field as again polls horizontally investigate began to due to a high cost US you can to a high cost to the followed by a lot of the you could do a lot less than the full of high cost the that you could do to low cost to the UK and so this is basically should be the You because of the cost of and his the result of the high cost of the falls subsequent reconstruction off the made this the details from each and you look at the low pasta the image of the city and edges for the average colours for whatever you are interested and budget
27:09
It basically looks like that itself for doing it is you take some image You down downtempo at Along the white exes with high cost of soap and a low cost but can't hypothesis and give you the details of the information that was changed The differences And as said love them on the road the of the White part Of this image can only where something happens there is a difference And these are the average this alone a care I've done it along the road to horizontal access X excesses Knowledge to would along the road collect so again by can do it a 2nd lowcost to Which was still give me a small a match Quetta the size of the original image and Wright and still containing many of the details because it's 2 times sampled the average and they get full possible this off Getting all the different These are the differences In the vertical reconstruction these are the differences In the horizontal reconstruction and these of the differences in both becomes a path to Times iPod Right This basically what I'm getting if I'd put together and these full images what do again The are original image Reconstruct perfect reconstruct I'm not throwing away any banned these sale images are in terms of resolution Exactly the original image It's just that Some of them are almost life That's very small differences very many 0 And that can seem like and get a good impression of a preview as he said some of the originally image to look a case What I'm saying this basic the worried do in in my resolution and and if we are doing it it looks like that of this allways and a double height and these although the difference at so the total number of each pixels in the status is saying that the Dior into the size of the image does not change it
30:25
At Paul than earlier doing all 4 combinations yes Again And what I basically doing it is you doing that white sell you go from the original image for 4 times the and 1 That will possible to be image for Load details sorry for for for basic feature of the new take this and again to the tools and the different combinations of the future then now with and even smaller and I could do that until you left with a single picks up and the common good and bad that Inglewood detailed information to the origin of fixed Icann schmooze reconstructed and image A Case this is basically a mighty resolution and a really care Osama full either It's actually so easy and a would be global food
31:39
Well during a so audio basically do we now was the I'm with the feature victory is used says the expected value and the standard of the Asian off the base of coefficient at each resolution to for example a few streets stage resolution you get 20 diamonds will feature of every because for every single 1 of these images is safe the spectacle you and the standard aviation so to numbers have turned off the field times to is 20 year 20 dimensional feature And this actually gives you a very good impression of the image that actually allows for very good Comparison of different images of I guess But The Because you need to be expected value and the standard of each exist to measure this is what the fact that uses to numbers your installed the expected value of which is a kind of like the average Colorado and that is that the average intensity but just having the average intensity is but to 0 2 0 2 2 mistake so you could have a lot of images having the same expected value but the and they might be different in the 2nd division yesterday might also be different askewness and and whatever it may be you now like so you could at moments to to distinguish between but the basic idea is the same as as we had with the greyscale images of all the texture mapping intellectus taking taking out statistical a moment to describe the actual image and this is where we do was the 1st to to statistical moment expected values stand defeat which we could add some help if we wanted to but to Aichi OK and a 20 dimensional feature does is is a very what the very easy to hand If which we would have 30 damaged but that it would take skewness would have plenty damage to the UK and the can Getting more more good that she brings to or 2nd public far too late shape based feature on the it it's really down to earth we want to see what in the picture we want to see all take we want to see shapes tables and chairs and you faces and everything and we have to find a way to describe it
34:36
Describes the shape using the picture information but not representing by picture information but by representing a something else we're with injury Thinking about its not so easy So I'm never less we have to think of a way out represent them because of the state system in which it really contribute significantly to the semi if we seek to images of caps 1 can be Brown when can be why we still notices simonetta to because the image contains and if we see the image of a tree or colour could be approached trio beach 3 0 0 3 0 1 of maybe the colours may slightly change But still we get the impression that similar and the impression is very off transport by the shape of things because it's just 3 shaped you know who you are Riesch and seeing that shaped everybody would agree yet to treat a and a one off Draw a camel something like that any more The fate last time with but that's the way it at that I'm So basically of also if you have shaped you may have died semantic information that this is the tree she Give you an idea of what the the author depicted in the show yes it could be a bunch of Colton The iPad on the state of its Tree shape but it's not treat but still something walks like a duck and talks like that it might well be the year that so is something shaped like lemon it might be time are now 1 of main and them
36:46
If we look at some typical shapes we find that the things come in and very many files for all these are immediately recognisable PM may be that not but the families are almost immediately recognisably chess though images utterly different that different coloured they differ in in Indian in texture firm here this woman texture you don't Or can they differ in Why actually in the shape of a basic shade For with seems to be there Club Kent And this is probably what makes us believe that such
37:39
And if you have a combination of simple shaped features of something from some round shaped some TriangularShapedCloud babies squid as and you end up other features before its of the colour of old text you could get a retrieval so for example if you around object in and out injured and red yellow which picture that might be Some said Or some of and with this kind of like Round Part and the rest of orange yellows order and that it might not best might be orange or the beach very well but I'm means it all we have and bomb
38:39
Identity fraud he was was 1 of the Ivy and to of looking for for cross shaped and of things There were quite well you know what you see that most of the match action to contain crosses and and even if they have enough like this the exact captain here or they split almost they will still be recognised as crosses and the something that can be done and that the fundamental problem of course is how do we recognise the shape of things to do and it was easy to 2 0 to construct something like added with the chasing a like is out the Apollo wasted on and he is the public lean to and these are the 3rd at the feet of the church for where they can do that easy for him on a flight some of the command the image and find out what is the the same but the computer to do that and in some automatic weight and of course as well as early as ideal semantic where mapping and it was below that suggested the shape of the shed
39:54
She has a certain shape because of has function And it doesn't make sense if you care upside down because he said all of that has to have a chat but the police true that from the shape of something you camp can actually OMX of the a computer wanted it is what was semantics behind thing and is not true that we have to have described the shape again with somebody to evict a with some load the features features and that the engine once we are able to describe shapes based on this descriptions we have to find simulator to measure To compel between different show is a table shape really different from a chest shape What are the sum error tease where it is the difference of a computer that if of this question the fundamental problem that we have to deal with the during next part of the and the 1st topic we will have it is segmentation a segmentations is really fundamental because not where the shapes that we know what our in the pictures all social on a picture of a topless in which is a Sunday on everybody can see that when it was time
41:31
While it's behind the drought It's not shown in the picture side don't have anything some shape Seat We knowledge that the picture does not show it it's all cluded It happened very off would only interesting shapes anyway Amin should you I'd should you help the look shaped They have about you Do we have felt you obviously large shaped Ways Something which she Which driving recalled where the interesting shapes But other the states that would make the picture seems symbolise to some of picture a pink So I'm during the consider all the different shapes to consider only be held on May Be only those in the 4th round of this is probably more of the image that it is a sundown image all than it is now image That problems were facing about what to do with them on the way to consider segmentation can be announced a problem because the the reprimand shade and the image and those not
43:13
Group The brush shaped We need The move Where and in review This unit The brash shaped Interesting concept the how to make out the images Because lots of the price and they are clued each other and they have tried in the UK and how to see what popular and 2 words the brought out of the city and and Brooke or mulch Where does this leave and Difficult the the same year remained the and and know what it is like them some rock full Meanchey but What of the shape of this kind of rope pupils into a new and more complex for we were brought up in the hype to make out her still yet of visual impression from its is that this is something in terms of shape and we can definitely see something The stop usual except 3 Blog It's all happening simply just for a complicated images and the Dome
44:52
The 3rd said 1 part is really all part of the state a book and sugar recording has been found and no with some minor with some usually of round the Sahaba livestock to adjust Take the knowledge and say yes but this is definitely a around some and if somebody look for for some always look for something round and to adjust to Stop this part Offers some so anybody can look for images whether some is how long The difficult part segmented of Kent
45:39
And by this segmentation as is 1 of the problems that the multimedia community is working on for 8 in all like and and an automatic SembCorp segmentation image segment of still kind of the Holy Grail of of computer vision and during off but maybe it has to be done automatically but but could be done semiautomatic that the kind of like it suggestions awful shapes that and images and just accept them a decline in the number of assistance is not and has been in in the early version of much media retrieval but it was it was not done automatically but there were some semiautomatic tools to forget that the use unit brought into the picture and the cost of the loss of some of us thought so for example was idea MPs cubic which is part of the beach to database the image extended but there is no way of doing automatic It's the way it is so idea that this this prototype thing you here and there you really have to go and an end to a new year low of tools like you probably no from from of British all PPL something like that of the of the last World of books kind of like that at just do something you had some some automatic of semiautomatic things with the kind of luck still some of the for most of the same colour some like to do that but they see the what you have to do this kind of have to draw the French
47:28
The adventure of again and you ended up with the different of and shapes and this for recorded the system as the Dutch
47:41
That time has said 1 of the of ways of doing it was kind of like I'm so called that feeling so if you had the early distinguishable regions in that nosy process rather things of of the same colour which could do with you could you could just Moskvy image and they all want a what the this this shape and and then fillet all the pixels was sustained colour and you would you would end up was a decent and the state could because of the image that either a it and someone on what from of surfaces and assume as you have something like here where the with a shape some of the broken it will immediately Run out and and this is no longer moonshake This is kind of like you know what I would seem more like that So there are made said that to the difficult and them many research project in multimedia retrieve that have been working on the top deck of the book unanimity just trying to figure out how to talk to a seat at the ground images here for example tigerish and but would see this as a time of not enough and that of enough and that sitting there for every image of the web and and like Annotating of the shapes the current shaped by hand I'm not some solutions somehow usually algorithms on the work 1 on special cases so if you have a special collection for example the of built on movies and the idea was to recognise people shape in in in movies and if you have something like that the and there is usually a way of doing segmentation automatically all at least Samuel with that of her behind the decision but it is fuel working on the Web or like general images which in possible it as he may be somebody things of level best difficult
50:04
So you to the segmentations problems actually from the commercial that basis shaped features Profile you remove So the prototype on the 1st set but I had them and the representation and compressed not shake used to enter the world of the segmentation problem made them unusable so they were removed from for 4 4 idea So the between 2 weeks and there Fouracre into media coverage and also for the IN for mixed that all decided to remove them There is an interesting point haul to do with the segmentation out to do the kind of like the PM the shape feature so we would cover them the lecture maybe some of the things that segmentation where or you a selection of images is of specific kind so you can have segmentation automatically and don't want to you to come to not know how to represent shaped features competitive that you should be at for that imprint put you can do is to find a film as the better perimeter of something the shape of a sail of something
51:31
The So the actual segmentations Ohio 3 get the old liable to get the sort of thing
51:45
You can have some nearest sell if you area has with similar brightness colour and texture of the meat may be long to the same of the of A reasonable if you have differences in brightness from 1 area to another them might be edge between the and edges over what is a good candidate for being part of Suez opened the visible caught at detection Of course You can fill spaces was more ethological operate doesn't just Mostyn would go into the later just see how I'm like wedded with a lot of here to see how far off some space extends to You may Sigmund the outlined as closed so you find the edges and put them together too close to and that might be the shape It also comes approximate it was some sort of political of the blind in early just interpolation and that is a large number of other procedures to to halt the rise and after more perfectly but they don't have complement each of the depending on called on what collection you have you might use that 1 on the other
53:13
But the 1st 1 is basically group show we are sure we do so in a small but not much Yes exactly so I'm Wrestling as basically 1 of the basic ideas The shape of the area That have the same right value Belong to the same cheques And but for wrestling icon distinguish different area was was different brightness was different intensity UK and some area that youngest wrestled and some very falls under stress and but it is Mostyn not images you will find that the of the ground images are light From them the background Well at least the different even if the dock all their black than the background life so photographing something in front of the similar came background and means not usually what he would do so are a certain stress old connect to separate between regions and what I do is basically you say that area that matically related like this Trulia celebrity value and can be separated from the surrounding space not like the
54:55
It is actually the best none in OMX and automatic men affect trading so qualitycontrol of part to try to find out something as an old not under computervision knows lot of these wrestling some of these wrestling algorithms and and we will just have a couple of what happens basically to take the brightest inflammation of the image And then you say OK output in the 1st told the Stop the bright area so like you And these are the black area like the screws over a And by just figuring out where the correct wrestled is icons Separated account from for ground and then icons
55:53
Do the shape segmentation Well I'm these ways is just take some wrestled up and this is the 1st of everything begun about 200 in brightness is background everything be For something and the case of fine images for easy to do but he's Eugene with great value images that show range of different colours and where was the difference is actually up their unnoticeable but you could free put your thing on whether this is a light at image as of Waldouck images level and the fixing the wrestled with all be not very good idea so we go for a flexible threshold that depends on the right about his to rhythm and tries to kind of like the idea what part of the histogram belonged to the full ground and belong to the back I'm usually the if you take his over his move that somehow all but you make sure you keep the peak to keep the masses of the distribution of and then you go looking for them for for the for
57:12
I'm 1 of the earlier algorithm that algorithms by Redland coloured which basically tried to divide the great values to Premier and a two part of any Calculate the expectation values of the great values in the Left and the Right so he separated in the middle and do you calculation of expectation value and not the new vessel should be the average between the 2 expected The show is the 1st told the Everett and recalculated expectation value The idea is that the expectation values in both part of the image Move to with 1 of the peaks Of the of the of the system and that once it reaches a peach that stabilise it doesn't move any more so the average dozens of any more than that basic when the algorithm finished up and the Saudi basically do is the biggest digram here Thinking the England You just had into 2 pubs become pupae expect Asian value of this part and of this part of the here and now The for granted but more new over her under the GNU recalculate The dilemma tel He look the loop look An cuticular In the middle of expectation values he recalculate expectations of UK easy to do said 1 is triangle algorithm some also about the same time which Connors's from where you should do it as you don't have a peak in the system gram Depending on whether the image is also for ground 0 Bakr of small images would have to be background peak Of the 4 Grand object have grumpy opened and wanting to do it is by a long queue connect the highest peak With the minimum In the distribution of Kent And the and iPad Pressault where there is a maximum perpendicular distance So basically out 1 to get at the foot of the peak And this is The where take the the threshold up care We connect the highest peak in the scramble side brightness value we maximised the distance to the connecting lines and distress sold is a minimum of 4 weeks shift of some comfort to the cost for so allow for some much look Also possibilities to kind of be termed of 1 of the 1st on the eyes of that is and iterative way of finding threshold this is just doing the solution for the perpendicular up Chris The he's right education example also and this is very often don't actually this wrestling is in medical segmentations if you have some of grammes for for example and want to find out something about of the baby's all can sell all different opens a you do every imaging Toulouse that kind of colour them you to the fact felt like which showed some of the stress early and using the extra his digram you you can find a way this facility so this is the interesting cost care and this is kind 1 of the background part can be removed from the image and and the but and the in what what actually happened was that was what should be can
1:01:50
That set up the 1st thing that also and a new set of rules and that the the value 1st on on different image area as we just as not just a 1 background and 1 3rd round of the image but there might be more Dixon the image the land together and I'll just but the images into different types and do settling on average
1:02:15
Can also because the bid and the image collection 1 of the order of nature of the images that would you get best and if you have the time and the best of reasons while the possibilities of especially
1:02:31
Doing doing Khaled images of so advise look at this in a two year segmentation works very well and I find all the flower beds segmented that they are kind of like you see here that these parts that are not really belonging to a sold out changes and colour you couldn't you couldn't do that with a lot but with toting you can't and found that that kind of had a very good segmentation that it out on the other hand as since he here with whatever it is I'm 1 of the things I'm the different types of talk about the light Politeia are broken down into different world order from public the animalness that they are more The all the other 1 is the body that you probably would like something on the amount that the SAS Chase and not really having all the different exotic and the lid of a tricky for the 5th try and sometimes for about them Aga Questions wrestling But Lamno
1:03:47
Aish tell you The present cost funded amazingly simple Discriminate for ground back his advantage yet to find the right vessels You suppose that the background and the for ground image read to change which is mobile withdrew from the and complex objects that have different pop like the loser can be di compiled by wrestling thinks that Bandirma with the picture of a 1 to have a break now 10 minutes
1:04:37
Get so that style along with the 2nd part of the front following was 1 thing but the eyes said that you could not only take that the area of the same colour but I could also figure out what the benches the limited of these areas and this is usually done in in X detection so I don't look at the area was a look at the limits of the law
1:05:14
And the beat the idea is that once you have closed around in some area and this is the shape of the and if you can constructed in such a way that semantically consistent objects us around by the the 1st perfect than you have read the silhouette of some of the and and what would you what basically do is you look at the bright function usually bid of what we already did when we were trying to figure out what Texas on so when we were figuring out what taxes are and how they can be considered but we looked at some the direct melody of part of the tumour mesh enough so we walked through the image in some direct and then so well here that it is light years it is likely that the stock had lied like is that the intensity function runs like lied and and its Dawkins than slide in the stock again no like that and the PM Looking at the intensity function as a real valued function you know you can order with look at the the derivatives and street where are extreme points of the function of all this point although he correspondents basically to this point of their because this is where of the documents and You came from light and you go back to life again finding this extreme points to maximise minimum in the function of you do that But I do like I didn't School in 0 1 over discussing the characteristics of of Look at the 1st derivative and look at all places where the derivative 0 because once you have the maximum minimum It's a point where the derivatives Sell This 0 And as soon as you move along with her The derivative gets nonfuel up Kent An interesting enough That different kinds of kerfs sold for example The because going like that A P yes For example This is 1 of them Also here you do have a point but the derivatives 0 but it's not a maximum them It just tell of Kent and the and how do you cheque whether to settle whether they are a real maximum momentum The book The the only ones That's right exactly city look at the 1st derivative It has the following after this point the same kind of kind of increased its either decreasing on both sides what increasing Mumbles but it doesn't change its sign What you need is something that really changes Negative to positive or the other way around for the next and the and the 1st 1 but they are 1st function is the gradient if you think TwoDimensional in terms of the function of this is the 1st derivative and the 2nd 1 of the last of Britain's 2nd derivative did describing the increase and the cream of the 1st of of the gradient
1:09:45
So what we are basically looking for we are looking for the gradient at some point And the problem was that all this is that it's a stepped function And I'd say it's not a continuous function because the Pixo has a certain intensity value and the neighbouring takes has a different intensity value and not that of his No no kind of like the way how they hold of the the by the transplanted of called for the kind of movie to into each other and but what you can do it you can't kind of like trying to estimated the differential functions from the point of you you consider the point the pixels cities and samples of function and then you kind of Interpolate the actual function for example for a transformation doesn't like that Then you just estimated causal for function for each picks from the immediate enable so you don't look at the function of the and image that once you have a picture You just say where I'm interested and the gradient of the pixel here in the middle of solo do is basically look at all the neighbouring pixels And just consider differences With respect to the pigs and then eye concede that the intensity got all down and this is all I'm interested and I'm not interested in the real function and and holidays in the UK and US interested is more or less and if by some some some intensity when going for the image in in any direct goals from from from brightness to adopt and up again to brightness then I'll know this is 1 of the point of I'm looking for an extra can see that Looking at the street pixels because of the year the intensity of 35 year to 0 in the eyes of some of the and only to reconstruct and if you do would like that are 1 of the 2 is that the so called gold said that it expected the estimated gradient just by looking at the the differences in the neighbourhood of usually the neighbourhood is a little bit bigger than just 1 takes Latino because you might have some some some voice that will give you a lot of holes positive but if you look at big enable you can really see by averaging opposite the right inside left side pixels of each Pixels full that you have to determine the gradient time you get a very good impression that this is an edge minimum local of a pet This was about as much faster than reconstructing the actual functioned based on the simple points but obviously there are some sold this is the way that we are up there
1:13:12
Some of the gradient based method 3 Calculate the magnitude of the gradient each point From And just show high gradient Because of a change in colour Area of the same colour showed no gradient at all just let going on 1 of that and then we use algorithm to separated the edges from the regions so it would be like that and we go through the image we find for example of the gradient here is not changing But suit as we get something The gradient has of the increase Oka And then This pointillist eating face Then it doesn't change much for the next time until you suddenly it goes into the background so we have a strong decrease the UK and and after that the change We require these 2 points and the 2 point of basically a point over here look by taking all these points from the match And fast thresholding the point was high gradient from the point with a low grade we will get quite except for UK Now that was a little bit suggested for because I'm using a by an array of kind of mine repayments here for a while about the EDS's where it's kind of like changing very little gradually into some of the more tricky is
1:15:04
So the actual advantage is that it is that it's amazing a simple to do it that way but the song as you have not assume as he slow gradient the something some area emerges into some of it does not work So if you merging Qantas over Qantas it does for because you don't have this Singlepoint where the gradient although some changes Sadly level we do what we look at the 2nd derivatives the and and look at it really crosses 0 was the derivative the 1st derivative the gradient really positive before 0 and negative after or negative before positive after If so what it went through some ballet all was amount If models It has changed somehow so the 0 crossing is really a test that shows you whether there was an extreme point a minimum all Maxim or not talk and and that this can be used and not the images way do have protected edges because of what see in a noisy amid even if you have time to go into the direction and have Excelsior even if you have a small changes of the intensity and so you have a five year any of the 3 year many of the earlier movie and the other by and what you can't see out its it it's not a big enough like change in what I do it very slow change even if you come to from 100 all something like that here The river is so change you can see that it's diminishing here And That's kind of taking up here So there must be 0 crossing Because people don't here and of the UK a This a basic idea But I saw what happened the idea is basically that if you have a blurred ApJ this you edged high intensity after words lowintensity before then agreed you will assess the seat and the go from here but you so you need 10 pixels to recover The differences in and in the level of function if you look at the the of the gradient The Brady in this kind of slowly of the gradient him gradient slowly rising as a function goes up At some point but the rope this is the interesting but this is where it changes start to full again so it that this is not the high his point in change that he would notice and that kind of like going down to get up and going up before And going down again and some point because of the the of and if you look at the at the last him filing below last year and is positive when the gradient goes up It is negative limit gradient those dump So it has to do the 0 crossing some Where does 0 crossing The maximum of gradient change And but this is where the action And though you have to be full An interval where the ECJ actually happened By looking at the gradient and the Boston 0 crossing you can put your point exactly where they should be UK and a new kind of like the gnawing the noise that is on both sides of But Yes but I'll like if Yes The Exactly Exactly so out 1 thing it is to really make sure that it is Gradient change that actually to a maximum and not just what changes some Holland and stays on the level of could also happened in which it is not an act United it's just a different different shift of shapes and it's just no like you have slightly different colours and the off because there is a shadow of what it means to be human but this rather slow and and even read changes goes to something totally different after which This is a sign that it probably is an act and you perfect I write the edge could be rather rather large and that the 2 2 2 2 0 2 0 a represents the state have to decide which the ex actually is account you can out of it for a really big modules for for the for the edges of the losses will be very difficult to compel between different shapes because of the dispute M'baye but some some white mountains you kebab between the UK that the real shape the real come to of the silhouette of the same could be within the mountain somewhere makes it very so deciding full behind has changed in the Grade II and it is this year its stake I'd like granted that but I'm in the best we can do
1:21:35
But there are For the heads to Texans is kind of like some interesting as dissent not to look only at the gradient change because the the new get these models to get a lot of 4th positive but that really 2 0 crossing of coffee and that gives you better come to us and we just Nault changes in colour of systematic and in the image I'm if you apply smoothing filter before calculating the Exeter then new even can lowdown Benoist for the you won't get a change in the Grade II and every every pixel that slightly this DiScala Also yes compound is only 0 crossings not the 0 points of what it could be just continuing change in terms of the gradient of the gradient for some reason to stop over 2 or 3 pixels some and a member and keeps on changes and this is not what we need as a This is what we need and we need to see they do with all pixels was 0 crossing and multiply them by the strength of the edge of the magnitude of the gradient and this is are While expected value that this really is the of certainty this is an extra and the actual segmentations and them be done by wrestling you keep all points that have been proven to be edges And everything the lowest certainty to just and I keep the 1 where you pretty certain that thing
1:23:33
So for example if you look at some images even with the There was noise We get some of the basic on to the for example of this is a strong edge also detected by solo filled up the same post here here Here again so I can see that the glue really only though the edges up covered by the cell and serve for this image works quite well but 50 at the face for example of the nose and that the debate messy and that the messy after the segmentation of wrestling to another Grammy we can every reconstruct than those of of the few points he of the confidence to be edges so for example we of this point here which which affects the edge she that it had raised does not help us and in the nose but it was very burglary and very very noisy and the origin of a majority if we have the right images it works This will try Good if a competitive ready and procedures and the 0 crossing technique of will only look at the the gradient because he that all these crossings said go to some point here for example
1:25:04
That is discovered that this Morrissey will also lead to the point In the gradient image But if we kind of look at 0 crossings to as 2nd indicated it that's really in edges there really is something else before and after and not just a means in in this case it obviously was light before it was light often the Justice wonder why not so this is not a change in the credit but he he It was like before it is not after this is a real change in the great was 0 crossing we can find out if we don't get the point if we just look at the Grade II and we will get these points and we will get the of points to the and Boise or
1:26:02
Which brings us to run extinct were so we can see the whole much lap of the book a so for the dog that there was an example of how to play with celebrity those fans crossing the and action it's quite simple but my club so 1st you need to transfer the in which from colour the reaching for greyscale or seeking which and you just have been a simple functioned edge is in the name of the function of the rest of the world in which the and intensity Belgrade doubled in each and the type of 31 by like in this case solo or 0 cross by the number
1:26:57
See how would be functions translates alive the view or what my club actually can and current
1:27:07
So why have we got here little between the Jews think you have already got a new school of these can be bought for about I'm just really very Jesus hygiene Belgrade defined Sinopoli's senior cost where they can use them in the Actionaid think cake Maybe with the money for something like this but the arrested
1:27:39
Built
1:27:42
Walk Poop The new And like a 150
1:28:04
Time It will that fought 12
1:28:21
Kennedy The mood Like inculpate some someone else
1:28:34
Like for example in the world with an IRA remove some information
1:28:50
This
1:28:53
OK so the for suspiration would be getting very Jews and pulled into the great Venice Well I've found hit also transformation of the reached lose this is what is needed for the head functions And then I'm just applying the front It functions to the images of my head of was nobody can read and will be So firm gap The 1st Dmitry Silverstone within the and the 2nd with a cross and it's actually quite easy to offer and that the show we were actually also happened
1:29:49
For a start but the should know from the whole If but
1:30:00
P
1:30:03
OK so there for the quality is not that great on the project with the way and they do greyscale you have been there in the right side of the screen and the in the left side is desirable
1:30:24
Well as you can see most of the edges of the would do a recognised you can see here the eyes of mild but the beautifully goldeneyes and the Bayern some modern some powerful gradient such you won't over his neck Lewis the head is somehow high recalled knives and told me to see how smoke slide for the 0 crossings And actually I
1:30:58
For my 1st surprise scene that the those is costing shoals of more details will recognise is a lot more about more jeans and the and the investigated to see why what Hepburn's he wants to about the reason and the 1st is the site implemented as functions where you can specify as that the threshold which he finds a gradient as an egg and those will feel the need for a few play with this sort of this gradient usually
1:31:34
What happens you can decide the amount Over the bill You can go go and It will quickly You don't want to be called Starting point gap
1:31:56
So why do so and the Treasury about the which gradient uses a gradient and a of pages are now but they did
1:32:06
The little church of my club actually has a punch and the whole book copulates distress of and made up with the with the threshold for each so it can be reaching the end of closer want more deepening crispiest official but not only ages and is that you will be with you get that actually via a this is where the boat is a low cost seems to the pounds a man cheques on the 2nd would be to see whether this is actually an edge or not
1:32:38
The broker also 1 of the featured which brings Up The
1:32:59
Me
1:33:06
Yet as you can see the play and the game is to be recalled nice of the somehow where it's rather below the rays of the sun sold as a way of the doesn't Radikal denies anything he denied claims 0 costing feel just to see if it will be able to go bust
1:33:26
The maker of this burden and their equal something in the City that the review would be much of a worry
1:33:36
Estimating the threshold
1:33:38
It is only able to your some and is also the
1:33:44
Over the blow so as you can see she says
1:33:51
This 1 is a better organize the song and you can play some will also and the total for the crossing the and get that would presentation even if part of the media bloating depends on where you can you want to to a for close but you are also able to get us roads with little costing characters
1:34:18
And for someone to get more the go collections can play with the with some within the skills the possibilities for the views of some sums might feel good decided 0 crossing and you can find all of them I have soldiers though in my club received 1 on the edge
1:34:41
All played is not On It everything will be saved so looking get also something like that the with the Royal Woods who though even though the based on the so you can anti differentiations between what happens when you win a prized almost goes to the House looking at by the Belarus and for or how came to get into his where they actually be more spread over the base
1:35:17
This was for the the 1st people get so that the next part is the of the transformation that allow you to do even find segmentation and a one off the road reknowned much less so called wash at transformation of the Shadwell basically award which is that part of the mountain I'm from where the World flows into 1 direction alongside into the other direct all the other side of for example is that of the all through Germany I'm from where the reverse stopped going down to the new and then to the Black Sea and on the other side of the bullish on the stock until Ryan River all of the of the revolt something and going to the mall to see like and this is the idea basically I'm the watershed information so shapes and area are considered to be based in that collect water and if you grow them they would grow together at some point and this is the watershed up and so on Every Great value has told of influence over pick as load of infrared where the intensity changes I'm not to much starting from some from some of the peaks of than actually the surface You will have this points where you meet of the of the stock from other pixels and this is where I put the water and the water is very good at proximation of an egg Basically the idea so of the GRAPE values are kind of like a typographical surfaces of the mountains Ulogie various was high grade values the public mountains and the area was lowgrade value as the ballets and if you mocked the distinguished points and start going from that you will need some this is where I should be so I'm worried that you do is you take a the minimum values minimum grave values of some area and then you start
1:37:58
The suspect that as we can see here nothing much changes The same year Here But then at a certain point you have to points And you go here and from the low point The you also go here they will meet And the change of the gradient But they have to overcome It's kind of a measure for this speech But I can't go so far it has to be using The spread between 2 different to and about what you get with a water that exactly shows this specific line of games Basically you large although pixels that have the same intensity as you pixels And once you go but not high although and intensity to that here to do that was the other pixels to and this is kind of like the level of water in your face of has to be the same level of pictures and at some point we will meet It was watershed This is a basic idea be between the behind water transformation and about what I do for a missed segmentation is the Dulecha transformation of the gradient so you look at the images
1:39:49
And you just see the intensity function through the imagined and you recall the gradient so you have learn gradient Olivia But once you get into this Blech He will have a high gradient OK if you inside the match you were below gradient up and nothing much changes you go idle submerged you The change in the grave and you have a high grade in and after that it will be of great Oka this basic to what you do they go from the or Italy image from gradient image and now aboue left church transformation in the image So while you do it is It takes The points Was the lowest gradient and you stop lobbying the area of the same trade A pin And assume that this Uma thought level In the intensity of the few enter the area or if you enter the area you have to do that for all the other base and 2 So also This can move up there And at some point they meet and this is what actually find such as the Lauch that Yes But and this segment the image by just footing the leftish at the separation limit on top of the origin of match This is what you get from a to and you can see that the burglary edges around here A kind of taking away and only become to of the water as the 1 possible way of different possible way of doing it
1:42:06
Advantage the enclosed and correct boring Hall and the point is we hobbies implemented efficiently because it is changing gradient and how do you kind of like film the base and is not really should have a right seat points and then iterative procedure to of your is kind of like and a greyscale of which can be pretty annoying and about what happens actually very often if you have to really images is that he stopped with very many points And since there was a transformational ways and when you reach the boundaries of some of point you can end up with Toriola Sigmund image The soothing good starting point for the full board of transformation is very MP and since starting point you the automatically over segmentation as 1 of the known problems of cause a
1:43:22
A 2nd possibility is active come to us by sell the the idea here is that regions are some kind of like the limited by the close of which the stadium boundaries and the close of could almost be a circle and of course acutely is not so so the way that the on that the image information of the intensity information along the circle changes This give you an idea of how show that should be the fold to actually get a good come to a certify stops As a single boundaries and you call it the snake because of snakes Italy into something and the inmates behind it like an image that is the type and he also and the and and you can see that become to a in the spot is really 1 of all the on boundaries but in this part of it is not The idea of the the active come to is that you minimize the energy of the snake were sold You have to kind of energy the internet and the G that is kind of like you wanted to be a circle and a stone as you have to be full made it as soon as you have to so because at some point I should of calls make it like this as if he had despite the loos energy use into and to the top for so to be default despite but you have the extra energy which is the image energy given by the gradient if the extra energy is very strong in places like yet where this really strong gradient the they've Comte Diethelm the Internet and But if the the Extern energy like here is very low because the gradient as they in the Than the defamation the internet and you can take over So you might well be segmentation problem as to energies working against each other to the Internet and the died trying to built the circle like ours and the extra oil energy pressing on these on the spread With the power of the gradient the stronger the gradient the more pressure fight with this kind of the for her as the basic basic idea behind the wheel you do is you stop with a circle and the and the internet and that she would try to broaden up the circles but as soon as you hit some area with a high gradient due out of energy push again said
1:46:44
At some point you don't manage so you you you pond and you want and you want some further but here you can expand any more because the external and tickets to strong stronger than the internet and from within the expunged and G I'm saying he had very high gradient so this spike in the in the curve can not be filled The internet and a tries to make it a circle tries to really smooth it But a bit like this but the extra energy is pressing down here were the gradient of the great change of the gradient and this kind of what is despite and then you end up finally at some probe that has a good curvature obsolete and G has tried everything to make it a circle to make some sort of some good shape and the external energy In all the places where it is very strong has the full circle of Yes well On board Put it in From The So you mean something like like a ring shape Fault Not We are all about Like that and that's kind of a spy Kia or something like that Because it it's account that it just come to a head It offers depends on the island on the extent of losses so if you of false is really of the kind It looks like that enough but very specific this nothing was in despite the and the internet and you can pressed into the spy and basic the again have the the balance of power from the extra energy prices back in all the direct suits But And the internet and you want to do a circus so it will probably not go all the way but it will look at the good size of probably you end up with something being slightly smaller than the other Figured that this balance of energy was does not read to various spiky can Yes
1:49:58
The balance the Well you could but of course that's solves sensible thing to do a mean you you you have to do it automatic for segmenting him And the and the of cost across the defined you get the active come to that is basically what you would end up with as segmentation And cost you probably don't want basic shapes you don't want very spiky think where every little kind like shadow all whatever it may be is taken into account but you on the basic basic shape of white use If you kind of have human shape want every details but you want from of like for of something for the head of the on and off the that could buy that it could set for something like that it was difficult to see will probably and that was a gingerbread man and that it but but still think this is the basic on to that you get because it has some good in terms of energy and a kind of like almost circa shaped and if there really was a person behind that you know what in the picture the changes and the gradient and put the pressure on the on the ethics and So I can do is you can you can also walk fuzzy edges because the kind of like a balancing the this energy and if the gradient it is slow and some area the curvature takes over makes it a nice growth slowed to send of the disadvantages of calls that with the acute accuracy of of the become so what you actually want from the point of the complexity of the crew increases and this of course means that the anti has to be calculated and different way Renesys with all with key but circle but if you want a more detailed shape saw the exact despite the shape of something like defining the in energy is this very difficult time and of course to have based his initial snake graffitti paint and then stops expanding but this is semi automatic rose so you draw about them than it itself to to the shape that you have the right helpful but you have to draw the match You should kabob get automatic
1:52:41
And that brings us to the last the to a for today's another way of from segmenting think left political operate to book a summer with a 2 1 pulled tool to statements in order to pull the pitch itself built up a big lead but this kind of operation might be difficult winning have noise or images the production environment when we say we have not and that this is an area where you have some kind of global shady delight doesn't the world and you have an example something like this Something during the growth of a new book that she becomes actually grey grey and you'll kind retailer that the difference where the the status of all start Gabon 1 and When something like this and you will stop of segment segmentation then what you come up with these through something like this But also with of Another of some of inside on your country globalised this scandal shapes saw what you actually want to to do is to make it to the states to called and so easy to to describe and in order to perform the took on the kind of applied preprocessing step that were claims they meet for modified selected for sort of has said his connexions here and this is what we political operate scandal
1:54:22
So the idea is that the political protest actually or operations that can be used on would so why don't they go only need but they look around a week so and will in the world by Richard overawed by regions and see how can actually part of the same and they can do this by Nebiolo would by either Heitinga pixels the week and of Qwest ahead the poorest below established what is my neighbour was the shape of my enable would and this usually used by in the case of more for Jacobo pointer by the the full 0 produced by socalled structures and lemon the basic operations director of political operations operators Campbell the Asian imagination has actually inflating the sofas so I'm making the surface but the good by the to the area
1:55:26
The opposite within the British so I'm shrinking the surface by removing some of the peaks And now we have to talk about the start of lemon sole the new board would what them playing this alterations and the plight of the most the use of such weapons selected the 11th hour of the across saw symmetry code This is the Pixelon coruscating and it's O 1 neighbour who is represented by use for big Or when a bowl with the sea in the early in a panic and so if you want looking in my new book would is given in the by benedictions outbound right here and considerable saw the
1:56:18
OK so the dedication and megapixels today in which the idea that he would here is that they my half as Matrix for convincing cities and I'm going each piece of with my new would she either the calls for both the disgraced and I'd by the Swede which beat 7 and look at my neighbour and see if you might be able or there is a peaks that the state should also be black That everything So that was a good 1 to increase the black area by looking game than a or beach but if I'd want a neglect weeks early in the new look would given by destructive adamant that it is based on 2 An unchanged and actually the effect of this mediation process is the 1 increased data create more break so say change way of exams and the other 1 is that and connecting snow objects or like those Mohoric you see me in in the 1st bought but
1:57:33
But be sure these on an example So why has he not only do not alone in which the given by Wild Gray Thesis here And what I'm doing in the in the Asian process is taking either 1 of these enable would change The structure the lemon and the need for Blue each The and that the plans were 1st Big settled up bound I'm getting out of the reach sold mapping interesting of a so called on the right they have a right but is not the a bad Pixelon or I am not willing to do anything that might have been a wide BigSim latest nothing going to happen here and sold by computer based in this direction and here and then each This big here But no choice Something happens when reached this BigSim Shia applying the class and the and when and non ICI that there is a great big so sought a back excelled And then a change this pieces here by adding need back into and leaking but this is the same also for these big so still Bombay have different being density being so and for this 1 and for although so that has been of the multi during the retreat pixels these the implicated but this is how my she moves What you see she that is the goal The and if I'd like the 2nd unstructured that amount for the same team which then I'd get some of the low inflation because it was the provision of the direction of the diagnosis just imagine taking this big so she The fight for the the structure of the lament squishy deny him and his direct Despite so this 1 becomes also L Is it Opaque P
2:00:03
Now the erosion is exactly the opposite I'd take again by the cell only this time of again with the ball Structured the limited 0 1 of them only against and I'm a big pixels wife Based on the same assumption of the new book will be a time 1 of BigSim and they will continue work would use their white if you then make mistakes also The effect of this is that the Games bought this appeal to have his that as a black peaks and you look around enable code and you see that happens there is a wide walking this BigSim should be also why The other effect is that you will it has its mobile in the wild exactly as we had with the ball but the bird
2:01:00
If we come back again to the Warrington million each and if you prefer the nation Then This BigSim guipure His appeal to be transformed into But again this happens also here But I should think Oh key What happens This context and she grew up with it place Job On the other side of turning the versions of the same Warrington of each It owns These 20 of the growth But it also destroys disconnections here because of the snow who it was origin of So if you conceded the preparation separately and it was kind of a problem because of 1 side you want to feed the cost for a want of perform with the Asian or on the other side He was the author of a new version and he 81 book such connexions Actually want no political operators is were found some kind of for each before the opening stand in the idea about openings is that you 1st before company golden
2:02:28
So you have both on occasions And then you prefer deletion to feel some of the costs for created by the leak are starting to to think that sort of of thing that in play You really can't take some by the fact that you have in the in each year or so away from the areas that the negating has been used the injustices what opening on the other side you can this the other way around the birth and 1st and inflation the nation and in the ocean The 1st scene Not once did joined the cost of date and pages
2:03:13
And this looks something like this If you prefer and when the 1st priority which Then the crews were 30 million of the same size so you for form 1st cloned erosion them the Asian You really couldn't do so connexion Shia But you still have a long history she On the other side by providing the closing the again you keep the size of the area impact you don't destroyed by 15 goals for the it will England youth of the world world's on her annual here and quote here but again You create some Symovonyk So actually
2:04:11
You have some of the big is the idea of using go somewhere political purpose for 4 0 imageprocessing is thought they would change but you can't for the sake man with of of the something like this for world 0 or see were whatever you do you can Amy and and use it on the New Year which but the disadvantage is that the degree of their use of the need to use your views is for more political all point they must be rather than the 40 have Johnson great will then have a problem you and used and there is not that I should have seen that the country is rather difficult to have blown world's and connexions then it that difficult to decide what they might point to by which sequence of about can a make sure that what comes out of this is actually a much better and can be can be can be easily segment sold in order to get the series precision the control difficult for for with more for culprit
2:05:24
And schemes where the and the lecture slower for the spoken about my is originally seeds we've seen no actually how we can be used in each of the country's allusions to every jingle Pixo so so clothing and pudding while the practical for pulling on the result of and then getting out of the smallest of the presentation of the reached the every 2 1 big still representing their peachcolored and the most important thing is that I'd want formations so that each there by 2 0 0 the difference so that they can requires the image that the biggest advantage is that the 1st contains a mobile to so is great for compressing impression that discussed about the a based features are the 1st possibilities using some kind of their based procedures were with spoken about the use of data And the young approach the idea here is that with 2 can now differentiate quiet cost between the programme in the background unity each just by choosing the dynamical thresholding the histogram in which the spoken about the edge detection above the solo Super and that the 0 crossing the both how we can use the time in the city the information in order of the day and using the regions and the spoken about some proposed a single of possibilities in order to clean up the region's for and the fact that of the other programs that these are boom or 44 or corridors next lectured next lecture will discuss about created by usually example will continue with the ship based features will discuss chained quotes full year description and a ball of them or complaints moment came very that for to think of 40
00:00
Random number
Pixel
Beat (acoustics)
Transformation (genetics)
Texture mapping
Multiplication sign
Computergenerated imagery
Characteristic polynomial
Heat transfer
Shape (magazine)
Measurement
Data model
Wave
Information retrieval
Frequency
Database
Energy level
Multimedia
Contrast (vision)
Feature space
Data structure
Subtraction
Descriptive statistics
Spacetime
Texture mapping
Information
Structural load
Bit
Measurement
Computer animation
Database
Multimedia
Pattern language
Energy level
Object (grammar)
02:01
Linear equation
Wavelet
Transformation (genetics)
Multiplication sign
Computergenerated imagery
Distribution (mathematics)
Calculation
Mathematical analysis
Heat transfer
Frequency
Equation
Pixel
Physical system
Algorithm
Wavelet
Information
Operator (mathematics)
Bit
Thresholding (image processing)
Variable (mathematics)
Mathematical morphology
Vector graphics
Functional (mathematics)
Shape (magazine)
Fourier transform
Maxima and minima
Wave
Computer animation
Vector space
System programming
Linearization
04:09
Wavelet
Computer file
Algorithm
Multiplication sign
Computergenerated imagery
Distribution (mathematics)
Mathematical analysis
Counting
Mereology
Computer icon
Coefficient
Mathematics
Term (mathematics)
Reduction of order
Representation (politics)
Information
Pixel
Subtraction
Linear map
Physical system
Covering space
Multiplication
Information
Computer
Structural load
Constructor (objectoriented programming)
Coma Berenices
Mereology
Transformation (genetics)
Computer animation
Graph coloring
Linearization
Game theory
Representation (politics)
Wavelet transform
Block (periodic table)
Resultant
09:10
Point (geometry)
Pixel
Transformation (genetics)
Multiplication sign
Computergenerated imagery
Mathematical analysis
Surface of revolution
Image resolution
Roundness (object)
Computer animation
Graph coloring
Raster graphics
Average
Singleprecision floatingpoint format
Process (computing)
Block (periodic table)
Pixel
Subtraction
Family
12:47
Personal identification number
Point (geometry)
Pixel
Correspondence (mathematics)
Computergenerated imagery
Gradient
Mathematical analysis
Set (mathematics)
Mass
Number
Template (C++)
Image resolution
Number
Arithmetic mean
Word
Computer animation
Arithmetic mean
Function (mathematics)
Representation (politics)
Object (grammar)
Pixel
Subtraction
16:33
Point (geometry)
Dataflow
Pixel
Musical ensemble
Observational study
Multiplication sign
Correspondence (mathematics)
Computergenerated imagery
Mathematical analysis
Average
Mereology
Number
Revision control
Average
Term (mathematics)
Data compression
Singleprecision floatingpoint format
Information
Pixel
Subtraction
Social class
Information
Fitness function
Local Group
Uniform boundedness principle
Computer animation
Raster graphics
Personal digital assistant
Data storage device
24:52
Digital filter
Mountain pass
Computergenerated imagery
Online help
Average
Mereology
Wavelet
Field (computer science)
Hypothesis
Frequency
Order (biology)
Average
Subtraction
Metropolitan area network
Musical ensemble
Information
Structural load
Physical law
Cartesian coordinate system
Componentbased software engineering
Wave
Digital signal processing
Frequency
Computer animation
Graph coloring
Figurate number
Resultant
27:08
Pixel
Multiplication sign
Combinational logic
Mathematical analysis
Mereology
Total S.A.
Emulation
Number
Hypothesis
Pointer (computer programming)
Video game
Insertion loss
Average
Term (mathematics)
Information
Pixel
Subtraction
Information
Structural load
Number
Computer animation
Personal digital assistant
Uniform resource name
Right angle
Matching (graph theory)
31:38
Standard deviation
Wavelet
State of matter
Texture mapping
Multiplication sign
Computergenerated imagery
Similarity (geometry)
Online help
Mathematical analysis
Shape (magazine)
Field (computer science)
Number
Information retrieval
Goodness of fit
Coefficient
Spherical cap
Authorization
Information
Subtraction
Rhombus
Physical system
Pairwise comparison
Standard deviation
Texture mapping
Information
Moment (mathematics)
Skewness
Division (mathematics)
Shape (magazine)
Table (information)
Similarity (geometry)
Computer animation
Graph coloring
Network topology
Contrast (vision)
Large eddy simulation
Coefficient
36:44
Computer chess
Rounding
Texture mapping
Computer file
Texture mapping
Computergenerated imagery
Combinational logic
Ellipse
Shape (magazine)
Mereology
Shape (magazine)
Information retrieval
Roundness (object)
Computer animation
Object (grammar)
Information retrieval
Order (biology)
Trapezoid
Family
Square number
Triangle
38:32
Group action
Texture mapping
Structural load
Multiplication sign
Computergenerated imagery
Shape (magazine)
Weight
Mereology
Computer
Semantics (computer science)
Functional (mathematics)
Shape (magazine)
Table (information)
Information retrieval
Summation
Computer simulation
Computer animation
Information retrieval
Identical particles
Subtraction
Error message
Matching (graph theory)
Fundamental theorem of algebra
Descriptive statistics
41:00
State of matter
Texture mapping
Computergenerated imagery
Shape (magazine)
Shape (magazine)
Local Group
Word
Roundness (object)
Computer animation
Term (mathematics)
Blog
Genetic programming
Units of measurement
Fundamental theorem of algebra
44:49
Rounding
State of matter
Machine vision
Computergenerated imagery
Insertion loss
Mereology
Shape (magazine)
Mereology
Computer
Number
Revision control
Information retrieval
Prototype
Roundness (object)
Computer animation
Personal digital assistant
Hypermedia
Database
Information retrieval
Revision control
Multimedia
Multimedia
Units of measurement
Row (database)
46:46
Surface
Pixel
State of matter
Decision theory
Multiplication sign
Computergenerated imagery
Denialofservice attack
Library catalog
Shape (magazine)
Web 2.0
Pointer (computer programming)
Energy level
Multimedia
output
Physical system
Adventure game
Dialect
Process (computing)
Online help
Projective plane
Shape (magazine)
Inclusion map
Computer animation
Personal digital assistant
Prototype
Form (programming)
48:45
Point (geometry)
Computergenerated imagery
Basis (linear algebra)
Set (mathematics)
Shape (magazine)
Shape (magazine)
Information retrieval
Prototype
Computer animation
Database
Hypermedia
Profil (magazine)
Personal digital assistant
Multimedia
Representation (politics)
Selectivity (electronic)
Perimeter
Oracle
50:49
Area
Polygon
Computer font
Interpolation
Texture mapping
Spacetime
Zeitdilatation
Texture mapping
Spline (mathematics)
Curve
Operator (mathematics)
Similarity (geometry)
Shape (magazine)
Mereology
Mathematical morphology
Number
Maxima and minima
Number
Computer animation
Quicksort
Procedural programming
Subtraction
Perimeter
Form (programming)
53:10
Area
Dialect
Spacetime
Computergenerated imagery
Range (statistics)
Shape (magazine)
Function (mathematics)
Thresholding (image processing)
Mereology
Local Group
Computer icon
Video game
Computer animation
Stress (mechanics)
Object (grammar)
Computervision
Right angle
Subtraction
Identical particles
Separation axiom
55:53
Histogram
Group action
Algorithm
Line (geometry)
Multiplication sign
Computergenerated imagery
Range (statistics)
Distribution (mathematics)
Binary code
Shape (magazine)
Mass
Average
Mereology
Distance
Thresholding (image processing)
Expected value
Maxima and minima
Average
Stress (mechanics)
Queue (abstract data type)
Energy level
Right angle
Subtraction
Library (computing)
Triangle
Physical system
Histogram
Algorithm
Shift operator
Prisoner's dilemma
Mereology
Line (geometry)
Thresholding (image processing)
Distance
Maxima and minima
Uniform boundedness principle
Calculation
Expected value
Loop (music)
Computer animation
Graph coloring
Personal digital assistant
Right angle
Object (grammar)
1:00:54
Area
Dataflow
Algorithm
Multiplication sign
Computergenerated imagery
Set (mathematics)
Thresholding (image processing)
Mereology
Rule of inference
Area
Roundness (object)
Mathematics
Computer animation
Average
Natural number
Order (biology)
Subtraction
Data type
1:03:44
Point (geometry)
Slide rule
Beat (acoustics)
Momentum
Hidden surface determination
Gradient
Real number
Direction (geometry)
Correspondence (mathematics)
Computergenerated imagery
Curve
1 (number)
Control flow
Limit (category theory)
Shape (magazine)
Mereology
Maxima and minima
Derivation (linguistics)
Sign (mathematics)
Mathematics
Video game
Term (mathematics)
Object (grammar)
Subtraction
Area
Polygon mesh
Physical law
Gradient
Complex (psychology)
Operator (mathematics)
Twodimensional space
Thresholding (image processing)
Limit (category theory)
Functional (mathematics)
Derivation (linguistics)
Computer animation
Mathematics
Helmholtz decomposition
Object (grammar)
1:09:45
Point (geometry)
Suite (music)
Neighbourhood (graph theory)
Digital filter
Musical ensemble
Pixel
Interpolation
Algorithm
Transformation (genetics)
Gradient
Direction (geometry)
Real number
Multiplication sign
Fourier series
Order of magnitude
Maxima and minima
Mathematics
Continuous function
Pixel
Subtraction
Position operator
Differentiable function
Area
Dialect
Algorithm
Point (geometry)
Gradient
Neighbourhood (graph theory)
Sampling (statistics)
Order of magnitude
Bit
Transformation (genetics)
Thresholding (image processing)
Functional (mathematics)
Data mining
Computer animation
Estimation
Function (mathematics)
Right angle
Uniform space
Differentiable function
Local ring
Matching (graph theory)
1:15:03
Point (geometry)
Digital filter
Pixel
State of matter
Gradient
Scientific modelling
Multiplication sign
Direction (geometry)
Computergenerated imagery
Insertion loss
Shape (magazine)
Disk readandwrite head
Order of magnitude
Maxima and minima
Derivation (linguistics)
Sign (mathematics)
Mathematics
Term (mathematics)
Core dump
Negative number
Active contour model
Energy level
Software testing
Noise
Acoustic shadow
Pixel
Subtraction
Multiplication
Area
Shift operator
Point (geometry)
Gradient
LaplaceOperator
Order of magnitude
Thresholding (image processing)
Limit (category theory)
Functional (mathematics)
Maxima and minima
Derivation (linguistics)
Position operator
Word
Computer animation
Graph coloring
Noise
1:23:32
Point (geometry)
Server (computing)
Musical ensemble
Mathematics
Computer animation
Confidence interval
Personal digital assistant
Gradient
Gradient
Noise
Procedural programming
Pairwise comparison
1:26:02
Scale (map)
Hidden surface determination
Digital filter
Group action
Musical ensemble
View (database)
Computergenerated imagery
Functional (mathematics)
Emulation
Hand fan
Number
Integral transform
Summation
Algebra
Computer animation
Oval
Personal digital assistant
Uniform resource name
Moving average
Physical law
Data type
1:27:42
Metropolitan area network
MUD
Information
Transformation (genetics)
Multiplication sign
Computer file
Cellular automaton
Grand Unified Theory
Hidden Markov model
Disk readandwrite head
Law of large numbers
Functional (mathematics)
Emulation
Maxima and minima
Computer animation
System on a chip
Uniform resource name
Radius
Units of measurement
Directed graph
1:29:43
Slide rule
Touchscreen
Projective plane
Gradient
Infinity
Disk readandwrite head
Thresholding (image processing)
Functional (mathematics)
Hand fan
Demoscene
Variance
Summation
Information retrieval
God
Finite element method
Computer animation
Commodore VIC20
Uniform resource name
Oval
Moving average
Website
Maximum likelihood
Quicksort
1:31:28
Point (geometry)
Web page
Gradient
Amsterdam Ordnance Datum
Thresholding (image processing)
Quantum state
Emulation
Summation
Pointer (computer programming)
Punched tape
Computer animation
System on a chip
Lie group
Personal area network
Game theory
Serviceoriented architecture
Simulation
Metropolitan area network
Supremum
1:33:06
Metropolitan area network
Presentation of a group
Differential (mechanical device)
View (database)
Total S.A.
Mereology
Thresholding (image processing)
Summation
Moment of inertia
Computer animation
Insertion loss
Hypermedia
Uniform resource name
RWE Dea
1:35:12
Point (geometry)
Surface
Digital filter
Pixel
Transformation (genetics)
Computergenerated imagery
Time zone
Water vapor
Shape (magazine)
Distance
Mereology
Integral transform
Maxima and minima
Oval
Moving average
Area
Scale (map)
Information
Surface
Structural load
Gradient
Transformation (genetics)
Computer animation
Oval
Units of measurement
Reverse engineering
1:37:44
Point (geometry)
Pixel
Transformation (genetics)
Gradient
Denialofservice attack
Water vapor
Latent heat
Flow separation
Mathematics
Energy level
Subtraction
Area
Gradient
Water vapor
Line (geometry)
Transformation (genetics)
Limit (category theory)
Measurement
Functional (mathematics)
Maxima and minima
Computer animation
Speech synthesis
Bildsegmentierung
Game theory
Matching (graph theory)
Separation axiom
1:42:06
Point (geometry)
Transformation (genetics)
Gradient
Computergenerated imagery
Curve
Mereology
Food energy
Power (physics)
Flow separation
Mathematics
Causality
Internetworking
Active contour model
Boundary value problem
Circle
Area
Curvature
Dialect
Information
Closed set
Gradient
Food energy
System call
Computer animation
Continuous function
IRIST
Right angle
Whiteboard
Procedural programming
Boundary value problem
Pressure
Data type
Active contour model
1:46:23
Point (geometry)
Complex (psychology)
Curvature
Multiplication sign
Curve
Insertion loss
Shape (magazine)
Disk readandwrite head
Food energy
Power (physics)
Goodness of fit
Mathematics
Internetworking
Term (mathematics)
Active contour model
Circle
Acoustic shadow
Extension (kinesiology)
Subtraction
Metropolitan area network
Area
Curve
Chemical equation
Gradient
Bit
Ring (mathematics)
System call
Arithmetic mean
Computer animation
Kolmogorov complexity
Quicksort
Whiteboard
Pressure
Active contour model
Matching (graph theory)
1:52:39
Pattern recognition
Surface
Neighbourhood (graph theory)
State of matter
Shape (magazine)
Mereology
Preprocessor
Object (grammar)
Operator (mathematics)
Active contour model
Noise
Pixel
Subtraction
Area
Operations research
Surface
Operator (mathematics)
Mathematical morphology
Shape (magazine)
Pointer (computer programming)
Computer animation
Integrated development environment
Personal digital assistant
Order (biology)
Noise
Statement (computer science)
Element (mathematics)
Quicksort
Uniform space
Data structure
1:55:07
Neighbourhood (graph theory)
Pixel
Zeitdilatation
Code
State of matter
Computergenerated imagery
Control flow
Area
Sound effect
Mathematics
Object (grammar)
Pixel
Area
Process (computing)
Surface
Open source
Operator (mathematics)
Sound effect
Mathematical morphology
System call
Element (mathematics)
Distance
Computer animation
Symmetry (physics)
Element (mathematics)
Right angle
Whiteboard
Uniform space
Game theory
Object (grammar)
1:57:33
Axiom of choice
Neighbourhood (graph theory)
Pixel
Zeitdilatation
Code
Direction (geometry)
Multiplication sign
Computergenerated imagery
Computer
Hypothesis
Sound effect
Flow separation
Mathematics
Population density
Integrated development environment
Data structure
Pixel
Social class
Multiplication
Process (computing)
Cellular automaton
Open source
Operator (mathematics)
Planning
Sound effect
Mathematical morphology
Element (mathematics)
Computer animation
Element (mathematics)
Right angle
Game theory
2:00:59
Area
Web page
Context awareness
Zeitdilatation
Operator (mathematics)
Mathematical morphology
Leak
Open set
Demoscene
Connected space
Revision control
Process (computing)
Gaussian elimination
Computer animation
Object (grammar)
Operator (mathematics)
Authorization
Negative number
Quicksort
2:03:12
Area
Point (geometry)
Series (mathematics)
View (database)
Image processing
Operator (mathematics)
Uniform convergence
Mathematical morphology
Degree (graph theory)
Computer animation
Order (biology)
Control theory
Metropolitan area network
Form (programming)
2:05:22
Computer programming
Query language
Numbering scheme
Histogram
Musical ensemble
Presentation of a group
Moment (mathematics)
Multiplication sign
Visual system
Boom (sailing)
1 (number)
Mathematical analysis
Thresholding (image processing)
Chaining
Programmer (hardware)
Chain
Invariant (mathematics)
Konturfindung
Descriptive statistics
Dialect
Inheritance (objectoriented programming)
Information
Moment (mathematics)
Operator (mathematics)
Thresholding (image processing)
Mathematical morphology
Shape (magazine)
Computer animation
Order (biology)
Code
Procedural programming
2:07:31
Lecture/Conference
Metadata
Formal Metadata
Title  Multiresolution Analysis, Form based Features, Thresholding, Edge Detection, Morphological Operators (28.04.2011) 
Title of Series  Multimedia Databases 
Part Number  4 
Number of Parts  14 
Author 
Balke, WolfTilo

Contributors 
Homoceanu, Silviu

License 
CC Attribution  NonCommercial 3.0 Germany: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal and noncommercial purpose as long as the work is attributed to the author in the manner specified by the author or licensor. 
DOI  10.5446/343 
Publisher  Technische Universität Braunschweig, Institut für Informationssysteme 
Release Date  2011 
Language  English 
Producer 
Technische Universität Braunschweig

Production Year  2011 
Production Place  Braunschweig 
Content Metadata
Subject Area  Information technology 
Abstract  In this course, we examine the aspects regarding building multimedia database systems and give an insight into the used techniques. The course deals with contentspecific retrieval of multimedia data. Basic issue is the efficient storage and subsequent retrieval of multimedia documents. The general structure of the course is:  Basic characteristics of multimedia databases  Evaluation of retrieval effectiveness, PrecisionRecall Analysis  Semantic content of imagecontent search  Image representation, lowlevel and highlevel features  Texture features, randomfield models  Audio formats, sampling, metadata  Thematic search within music tracks  Query formulation in music databases  Media representation for video  Frame / Shot Detection, Event Detection  Video segmentation and video summarization  Video Indexing, MPEG7  Extraction of lowand highlevel features  Integration of features and efficient similarity comparison  Indexing over inverted file index, indexing Gemini, R * trees 