0:00:14 | you best in my on the body |
---|
0:00:17 | got information with the one on you but still |
---|
0:00:20 | speak |
---|
0:00:21 | also uh the company |
---|
0:00:25 | so i would like |
---|
0:00:26 | to a a kind of my to work |
---|
0:00:28 | so so is the |
---|
0:00:29 | T G minor but a |
---|
0:00:30 | when you can use |
---|
0:00:31 | yeah nice |
---|
0:00:35 | problem of minutes before starting to what |
---|
0:00:37 | yeah |
---|
0:00:38 | we can start five it's early but don't want that for |
---|
0:00:41 | start that for once one the ways people coming in |
---|
0:00:44 | maybe few so what a problem but minutes |
---|
0:00:46 | for |
---|
0:00:47 | but |
---|
0:00:58 | one one question was really useful for a single all you here are present the string |
---|
0:01:03 | oh the any or consider here from the or |
---|
0:01:07 | uh_huh but here's for marketing |
---|
0:01:10 | what fraction of room |
---|
0:01:17 | rest of five db |
---|
0:01:22 | well the question is that i can use the reading in terms of like S it's all of using the |
---|
0:01:27 | more robust tree |
---|
0:01:28 | but to be the mouse trying to paul |
---|
0:01:31 | what a things that doing is thing to think these of like guess |
---|
0:01:34 | there are only a you can be used |
---|
0:01:36 | feedback to get results way |
---|
0:01:40 | or |
---|
0:01:45 | i this base or remote start |
---|
0:01:48 | so |
---|
0:01:49 | right |
---|
0:01:50 | these |
---|
0:01:58 | i present proxy |
---|
0:01:59 | the method |
---|
0:02:00 | like in that |
---|
0:02:03 | i have divided my presentation |
---|
0:02:05 | for by |
---|
0:02:06 | first of it you you of hmmm you of their scope i |
---|
0:02:10 | i with that |
---|
0:02:11 | and you know |
---|
0:02:13 | you convex function minimize on the |
---|
0:02:16 | complex |
---|
0:02:17 | modeling right |
---|
0:02:19 | and of so that a |
---|
0:02:20 | that my |
---|
0:02:21 | me |
---|
0:02:22 | then um of different the uh |
---|
0:02:24 | the |
---|
0:02:25 | they are either with a |
---|
0:02:27 | one of the most competitive math |
---|
0:02:29 | based on |
---|
0:02:30 | a great for section |
---|
0:02:33 | and |
---|
0:02:33 | hmmm one on my proximity |
---|
0:02:35 | a technique |
---|
0:02:36 | based on proximity operator |
---|
0:02:38 | computing |
---|
0:02:39 | i |
---|
0:02:40 | project |
---|
0:02:41 | find a yeah present some results |
---|
0:02:44 | as a as the issue of of but it just i would that estimated very at |
---|
0:02:50 | like in uh |
---|
0:02:51 | the am |
---|
0:02:54 | so let's start with that this of the disparity |
---|
0:02:58 | given to uh |
---|
0:03:00 | to images |
---|
0:03:01 | the in from |
---|
0:03:02 | two different point |
---|
0:03:03 | of you |
---|
0:03:04 | we can see that |
---|
0:03:05 | the the so that that and that's is is presented by court |
---|
0:03:10 | why for example |
---|
0:03:12 | and uh uh is and the right image with |
---|
0:03:15 | uh |
---|
0:03:15 | X with different |
---|
0:03:16 | so we will uh not as |
---|
0:03:18 | by |
---|
0:03:19 | right |
---|
0:03:19 | my |
---|
0:03:21 | we define a is simply |
---|
0:03:23 | and this is between these to me |
---|
0:03:25 | so you will be |
---|
0:03:27 | uh this is a T |
---|
0:03:29 | and X prime |
---|
0:03:30 | why might white |
---|
0:03:32 | or or or of its |
---|
0:03:33 | a rectified |
---|
0:03:35 | so why |
---|
0:03:36 | Y |
---|
0:03:38 | but it is a is to be a |
---|
0:03:43 | oh problem to estimate a is is the main study |
---|
0:03:47 | so we ate to fine |
---|
0:03:49 | the corresponding pixel |
---|
0:03:51 | and the last |
---|
0:03:52 | i |
---|
0:03:54 | the state of the art the can find might yeah approach |
---|
0:03:59 | but in that we can uh can not mentioned |
---|
0:04:01 | feature matching approach where a with ace |
---|
0:04:04 | yeah |
---|
0:04:05 | curves |
---|
0:04:05 | and segments from the images |
---|
0:04:07 | also global methods |
---|
0:04:09 | as the dynamic programming and evaluation an approach where they are mean you at they minimize a global the energy |
---|
0:04:16 | from |
---|
0:04:17 | also some fines that use a of the cross or local correlation we those |
---|
0:04:22 | as a normalized cross-correlation correlation |
---|
0:04:27 | in want to follow we will a light by less a variational methods where our cost function would be uh |
---|
0:04:33 | and then are measure between |
---|
0:04:35 | the left image |
---|
0:04:36 | and that right image compensated by the despite its you |
---|
0:04:40 | and the |
---|
0:04:41 | we will use |
---|
0:04:42 | and then are or five |
---|
0:04:44 | that is the proper a lower semi-continuous convex function |
---|
0:04:49 | and the sum of the as this or will be on them |
---|
0:04:51 | on our limits it's the court |
---|
0:04:54 | but is this function is convex with respect to the disparity you |
---|
0:04:58 | so to avoid a non-convex optimize addition |
---|
0:05:02 | a problem |
---|
0:05:03 | we will uh suppose that the uh right image compensated by a by to you is footage |
---|
0:05:09 | we will consider the first taylor expansion of this not than the two |
---|
0:05:14 | and uh uh a uh a and fact we |
---|
0:05:17 | a compute this |
---|
0:05:18 | first or expansion |
---|
0:05:19 | and the first to make you but |
---|
0:05:22 | so i right image compensated by a but to you would be it well |
---|
0:05:27 | to the right image compensated by a and the of value |
---|
0:05:30 | you by mine is you might is to bottom to light |
---|
0:05:33 | by the or something to do and and of the disparity compensated right |
---|
0:05:39 | so to simplify the notation we will consider as D is the gradient |
---|
0:05:43 | R is the right image compensated by a it's paid to you bought class you bar T as the left |
---|
0:05:53 | so our cost function will be at most of five which is convex function |
---|
0:05:58 | function of the you minus R |
---|
0:06:01 | the some always is a on the support them at |
---|
0:06:05 | but |
---|
0:06:05 | as a with few it uh if we aim to search that's very to you that minimize i would criterion |
---|
0:06:12 | and this case that this problem is and it's pose problem |
---|
0:06:16 | because we can find uh a and and a finite in fine the sole use some of this that convex |
---|
0:06:22 | so we will allow uh and |
---|
0:06:24 | some convex constraint modeling prior knowledge and observe that that a of of our is to me |
---|
0:06:32 | that mean we will present our problem as a set estimating |
---|
0:06:36 | set theoretic estimating a problem |
---|
0:06:39 | that's mean we with search i would this to you |
---|
0:06:42 | that minimize our criterion J |
---|
0:06:44 | and to don't to then there's section of different constraint that we also there |
---|
0:06:51 | so |
---|
0:06:51 | what's can be these constraint |
---|
0:06:56 | it can be done and looks about the uh minimum and maximum amplitude of the disparity you |
---|
0:07:02 | or all that are also it's can be the uh uh uh you can produce an upper bound on the |
---|
0:07:08 | total variation of i would spend at mouth |
---|
0:07:11 | and this constraint to you get not does that that it's a piece there of the discontinuities and smooth some |
---|
0:07:15 | was it almost in is ideas |
---|
0:07:18 | or or can be a cost with that constraint |
---|
0:07:21 | we can introduce use an upper bound also on the norm and one norm of the weight of coefficient wavelet |
---|
0:07:26 | coefficients of our despite |
---|
0:07:34 | so to summarise we are minimizing a compact |
---|
0:07:37 | criterion J under different convex constraint |
---|
0:07:41 | as i said before that the sub the get projects is one method |
---|
0:07:45 | it's would used and uh before that two thousand and nine by me that |
---|
0:07:49 | and the |
---|
0:07:50 | uh use a fact |
---|
0:07:52 | two |
---|
0:07:53 | so this problem |
---|
0:07:54 | it's that convex what front |
---|
0:07:58 | to uh because of the algorithm that used and also they were lies to add |
---|
0:08:02 | convex "'cause" the convex there |
---|
0:08:05 | which is as far you might you box way |
---|
0:08:09 | so they they an is that what that the can a convex |
---|
0:08:12 | uh criterion |
---|
0:08:14 | and also they yeah uh so we use subgradient projections of the this to you on the different convex constraint |
---|
0:08:20 | that they can there |
---|
0:08:23 | this this minimizing is a on the uh number of close and are yeah |
---|
0:08:28 | uh |
---|
0:08:29 | and the two images left and right one that's mean we can that here just the peaks still that up |
---|
0:08:34 | here in the last |
---|
0:08:35 | and that i |
---|
0:08:38 | not though we will so our problem that is a convex function J |
---|
0:08:42 | and the different combat |
---|
0:08:44 | a that we consider |
---|
0:08:45 | but we are not hold lies to use just |
---|
0:08:48 | sickly convex quadratic function |
---|
0:08:50 | we have a a great flexibility and that's source |
---|
0:08:52 | of this criterion |
---|
0:08:54 | so to introduce i would want as the definition of |
---|
0:08:57 | that the proximity operator |
---|
0:08:59 | first |
---|
0:09:01 | if you have a points why |
---|
0:09:02 | that's you wanna project it on a convex set C |
---|
0:09:06 | uh we can find this projection |
---|
0:09:08 | by minimizing the come at the problem had to get a function of this constraint |
---|
0:09:13 | plus the uh |
---|
0:09:15 | quite that take distance between a |
---|
0:09:16 | and what |
---|
0:09:19 | so if we saw but this problem we can find the projection of the point why on the convex constraint |
---|
0:09:27 | for a place them to get a function by and that and that but three function F |
---|
0:09:32 | so our problem would be minimizing the function F plus the the distance between a |
---|
0:09:38 | and what |
---|
0:09:39 | this all use of this minimization problem will be the proximity operator of the function F at point point |
---|
0:09:51 | just to uh |
---|
0:09:53 | uh uh rewrite our problem and different way so we are minimizing our convex function J |
---|
0:09:59 | on the different convex comp constraint |
---|
0:10:01 | that can be modelled it it's by a line or operator and time |
---|
0:10:06 | the minimization is on the non on clothes and an are yeah |
---|
0:10:09 | and use the the error measure of five which is convex |
---|
0:10:13 | a function of to you minus R |
---|
0:10:16 | how we can solve that |
---|
0:10:17 | use a a to here i will present |
---|
0:10:19 | but are are good as a |
---|
0:10:22 | uh a which have especially specially will be a the P B X eight plus and yeah i go to |
---|
0:10:27 | them |
---|
0:10:27 | which i low as to minimize this context can at a complex function |
---|
0:10:32 | uh was some closed convex constraint C i |
---|
0:10:35 | uh just |
---|
0:10:36 | if we are able to compute the proximity operator of the criterion J |
---|
0:10:41 | and |
---|
0:10:42 | the project directory |
---|
0:10:44 | would what estimate on different constraints C i so our problem will be solved |
---|
0:10:49 | so simple |
---|
0:10:50 | it X a plus algorithm |
---|
0:10:52 | so |
---|
0:10:53 | we can a our algorithm was some way |
---|
0:10:57 | uh related to the uh convex constraint that you have |
---|
0:11:01 | and also go a a positive constant related to the of does zero point to the cost function that we |
---|
0:11:07 | have |
---|
0:11:09 | we initiate our our mattresses this |
---|
0:11:12 | and we compute the factor you which is and then G of different up a or operator that so uh |
---|
0:11:18 | we have a D complex cost |
---|
0:11:21 | as we see in that in each iteration and our there is a |
---|
0:11:25 | we project |
---|
0:11:26 | each month is on to convex |
---|
0:11:28 | a |
---|
0:11:30 | and also we compute the proximity operator of the criterion J divided by calm |
---|
0:11:37 | so we have our just that are i and i would this but T U |
---|
0:11:41 | by relaxation parameter lump |
---|
0:11:44 | so the most important than the side them that's we compute the proximity operator of the criterion J and we |
---|
0:11:49 | have |
---|
0:11:50 | X it's form for different convex a |
---|
0:11:53 | a cost function |
---|
0:11:55 | and the guy died projection on that the D convex a constraint that the i is not so uh so |
---|
0:12:03 | so to uh |
---|
0:12:05 | to improve the performance of this and good as a mall present some results |
---|
0:12:10 | the first one a corresponding to the peer what it off |
---|
0:12:13 | he how we present the left image and the related to going to tools |
---|
0:12:17 | and i present some generated disparity map using different method to can but to compare it to our method |
---|
0:12:23 | the first one is the block |
---|
0:12:25 | despite the estimation |
---|
0:12:28 | this a one that that i presented before using got to the what that the convex function |
---|
0:12:34 | based on sub granting projection |
---|
0:12:36 | and the our but the X a plus and go to them and we choose that of the and one |
---|
0:12:41 | norm |
---|
0:12:41 | and a a and that are much in our would cost of from |
---|
0:12:46 | also we compute the psnr between that is generated this to my and the ground truth |
---|
0:12:51 | as we see that our method gives the best result |
---|
0:12:55 | and some important what improvement and the disparity map that you have |
---|
0:13:02 | and of the results on the teddy pair |
---|
0:13:05 | so the last image it's and the ground the true and the generated disparity map using the same method |
---|
0:13:11 | oh i will have a real be the B D easy to it's of block that despite the estimation and |
---|
0:13:17 | the L two norm that is |
---|
0:13:19 | strictly convex function |
---|
0:13:21 | and and one norm based on the X eight but i |
---|
0:13:26 | as an application of the uh estimated disparity map will present |
---|
0:13:30 | uh that's that uh its application and stating image coding |
---|
0:13:35 | as we know that and uh to compare the method with an but uh that you image coding a we |
---|
0:13:40 | um |
---|
0:13:41 | codes first |
---|
0:13:42 | uh the last |
---|
0:13:43 | and the right image independently |
---|
0:13:45 | and here we apply the same at loss for on the left and right image using find three |
---|
0:13:50 | we have that like possible |
---|
0:13:54 | many works are done in a stereo image coding using john coding scheme |
---|
0:14:00 | uh based on a compensation in uh |
---|
0:14:03 | but |
---|
0:14:03 | the disparity a compensation this right image |
---|
0:14:06 | so |
---|
0:14:07 | at consist of code one image you which choose the right one |
---|
0:14:11 | and a residual image |
---|
0:14:13 | here here uh can define with the difference between the left image and the right image compensated by the despite |
---|
0:14:19 | its you |
---|
0:14:20 | and all the the the disparity was scored |
---|
0:14:25 | so |
---|
0:14:25 | we have to image to to the right |
---|
0:14:28 | the is it image and that is but |
---|
0:14:31 | and you were compared our we you method the block that |
---|
0:14:34 | but the if estimation that is by estimation |
---|
0:14:38 | using |
---|
0:14:38 | strictly a complex what the tick function an |
---|
0:14:40 | and one or based on the X eight plus and to |
---|
0:14:45 | the resulting wavelet coefficients are input in uh i i are encoded use it uh using a you back to |
---|
0:14:50 | thousand |
---|
0:14:51 | and also as a dense fields are and that the by applying a what be the compositions |
---|
0:14:56 | for the by an entropy coding with at |
---|
0:14:59 | point to six four |
---|
0:15:03 | so uh oh here we present the psnr of D reconstructed images |
---|
0:15:09 | left and the right one |
---|
0:15:11 | versus the bit error rate |
---|
0:15:13 | as we see that the it depends |
---|
0:15:15 | scheme |
---|
0:15:17 | uh give is of the the last P high |
---|
0:15:20 | and a couple states |
---|
0:15:22 | compensation this by the estimation |
---|
0:15:24 | here are give the best to all that to me in that john coding scheme uh give that uh the |
---|
0:15:29 | great yes and |
---|
0:15:31 | and you don't the L T P S and are given by our method presented by the red curve |
---|
0:15:36 | does this by the estimation using the and one norm and |
---|
0:15:39 | and then that the estimation do think this strictly convex function with it's which is whether T |
---|
0:15:44 | and a lot that estimation |
---|
0:15:50 | as i said before that our criterion J is complex |
---|
0:15:53 | and we have |
---|
0:15:54 | am a great flexibility and the choice of our i don't mess |
---|
0:15:58 | so we can use this so |
---|
0:16:00 | but |
---|
0:16:01 | to uh uh the case one our image images present some noise change |
---|
0:16:06 | that's me |
---|
0:16:08 | oh |
---|
0:16:09 | if you have a a a a for example so that paper |
---|
0:16:12 | oh |
---|
0:16:13 | the we the guide that it uh the court it or P with all that people are so we can |
---|
0:16:17 | use and one norm |
---|
0:16:18 | or or when we have a what's and noise we can use that this |
---|
0:16:21 | could but this that's for example |
---|
0:16:23 | because of when we have a a noise uh |
---|
0:16:26 | we we can get the prove that uh could but this does that can be used to uh |
---|
0:16:32 | to make a less this effect of patient |
---|
0:16:35 | so here we compare the at sort or to uh to D X i plus using and one or when |
---|
0:16:41 | we have a a so that paper |
---|
0:16:43 | noise changes |
---|
0:16:44 | and you can see that the uh and i where is all the the uh noise as the |
---|
0:16:50 | we don't have a is that in fact |
---|
0:16:52 | and that and when we have a a a noise |
---|
0:16:55 | the uh |
---|
0:16:56 | disparity map but |
---|
0:16:58 | present |
---|
0:16:59 | morse smooth are yeah and the P of this company D used for the objects that in that present in |
---|
0:17:04 | the uh a and the that |
---|
0:17:05 | disparity image |
---|
0:17:10 | and the some uh very numerical result using the snr and absolute you'd error measure or are given and that's |
---|
0:17:16 | prove that our method |
---|
0:17:18 | gives a uh are provide an a create that's uh map |
---|
0:17:24 | to conclude uh |
---|
0:17:26 | i would like to uh |
---|
0:17:27 | some right so |
---|
0:17:30 | we form our of our problem as a convex function |
---|
0:17:34 | that we are minimizing under different convex can |
---|
0:17:39 | and uh |
---|
0:17:40 | we have a uh are at that is to choose the error measure as we want |
---|
0:17:45 | so that we have various criteria |
---|
0:17:47 | and also just |
---|
0:17:48 | if we are able to compute the proximity operator of the criterion |
---|
0:17:53 | and also a make |
---|
0:17:54 | direct projections of the estimate on different convex constraint |
---|
0:17:58 | so we can run the algorithm simply |
---|
0:18:00 | and also i present two applications |
---|
0:18:03 | and the presence of noise |
---|
0:18:05 | uh when and the left and the right image and also on the is a uh and state you image |
---|
0:18:10 | coding |
---|
0:18:14 | a what i'm working on a this time is that to a a a uh i is then extending my |
---|
0:18:19 | yeah my and good as them to the case when i have a nation variation |
---|
0:18:23 | and also i would like to applied to a for the colour image |
---|
0:18:28 | that is that thank you for that then |
---|
0:18:37 | still have a questions |
---|
0:18:39 | it's really here |
---|
0:18:40 | but |
---|
0:18:47 | so that projections to compute |
---|
0:18:49 | or or a non be or numerically |
---|
0:18:52 | and the better |
---|
0:18:53 | remote yeah so what was that the traditional load of the scheme as compared to the mid those that are |
---|
0:18:59 | present |
---|
0:19:01 | uh competing to the of getting projection |
---|
0:19:03 | you have a an the other and yeah |
---|
0:19:07 | uh |
---|
0:19:07 | a fact in the lot |
---|
0:19:09 | by the estimation uh so |
---|
0:19:13 | log and the fact that based on a block disparity is so they uh compute block and the |
---|
0:19:18 | there are but |
---|
0:19:19 | in the in uh to get in projection that are obliged to compute the this again at an operator or |
---|
0:19:25 | or the uh |
---|
0:19:26 | so when we use when that use for example of the of to variation there are a like to compute |
---|
0:19:30 | the sub gradient |
---|
0:19:32 | difference of of the total variation and applied the projection the complex |
---|
0:19:36 | but no no method in fact so we use the |
---|
0:19:39 | uh a applied the projection of the operator on the comp |
---|
0:19:44 | or what loss in computational complexity |
---|
0:19:47 | how much computation how much time does it big |
---|
0:19:50 | run |
---|
0:19:51 | these different all tomatoes all comparable the |
---|
0:19:54 | what for |
---|
0:19:56 | and fact we a plot we are caught a we program this algorithm on matlab of software |
---|
0:20:01 | and now we are applying and do you P G P U because uh as we see that and the |
---|
0:20:06 | algorithm we have a the ability to light |
---|
0:20:09 | a a set of |
---|
0:20:10 | a D projection we can uh |
---|
0:20:11 | projected that the but and but i don't way and also the uh proximity operator can be uh computed and |
---|
0:20:17 | but ah |
---|
0:20:18 | so we are waiting the fact the results of because uh the work at the |
---|
0:20:23 | always the |
---|
0:20:24 | a that are working on |
---|
0:20:29 | of the questions |
---|
0:20:36 | you |
---|
0:20:44 | uh |
---|
0:20:45 | question uh and fact that we are going to the uh sub getting projection technique |
---|
0:20:51 | and it was proved that |
---|
0:20:52 | two thousand and nine at the best that technique get pop to the other methods |
---|
0:20:56 | so we are made of our uh given there is not better than the best techniques so |
---|
0:21:04 | oh the question |
---|
0:21:09 | but at let's thing we once again |
---|
0:21:12 | i |
---|