0:00:14yeah to go and
0:00:17the paper in the thing pretty
0:00:19with flat
0:00:20i image
0:00:21the presentation divide it will be defined
0:00:25for as we we will give motivation then uh a very brief
0:00:30but know
0:00:32then we you are going to review the the frame pretty each that's proposed for each the
0:00:38the difficult
0:00:39to to for using this flat transform
0:00:41the proposed solution and modification the
0:00:45model selection
0:00:46after a going
0:00:49it's some experimental results and then conclusion
0:00:54the first question must be white to use that the transform
0:00:58when most of the
0:01:01in like to a specially if you do all standard
0:01:05that's a one
0:01:06use plot and so
0:01:10the the fact that of using that and some since this a part of this used just a you in
0:01:15the block itself
0:01:17maybe to a
0:01:18that would be good read the shown
0:01:20the block yeah so
0:01:22besides that
0:01:24for the same reason we can exploit
0:01:26had to the sum of neighbour block
0:01:29that was before you know
0:01:31both of those
0:01:32the leads to a superior performance fine coding
0:01:36in objectivity in subject to
0:01:39and measure
0:01:41but that comes list the expense of for higher combat
0:01:45and this point
0:01:46the only as
0:01:48uh you might to coding standard that uses
0:01:50let but are song it i is J back so
0:01:53that was proposed but microsoft corporation it was
0:01:57formally known
0:01:58i those for
0:02:01you from go it
0:02:02uh have proposed a a new interest frame prediction it not then
0:02:06and we propose here we propose a inter-frame prediction in the pixel domain
0:02:11much seem not what's then each thus
0:02:16no we'll go to a very brief
0:02:21uh you know that transforms at in block transform this you know is divided in block
0:02:27and that will be processed the in
0:02:30but different from
0:02:31block transform
0:02:34uh this part will be bigger
0:02:37we have a a a a and this new row that will have in the time more samples then
0:02:42the original one that would call
0:02:45to a different from the traditional X
0:02:50and is the but as the this
0:02:56as a that it will have been times more
0:03:00samples was
0:03:00this and will be called overlap set
0:03:03and we have a here for that the example for the overlapping factor of two
0:03:08we can see that half of the sample
0:03:11but an important blocks
0:03:13in the last and the
0:03:15will be used
0:03:16the of the traditional samples that were or a red to use it in the back
0:03:23B B and Q
0:03:24the matches is contained in the direct and inverse transform the sound will be held
0:03:31in the same way as in black dots on
0:03:33with the difference and uh
0:03:35and that
0:03:36the thing yeah
0:03:39will be different from the original one
0:03:43where that because the traditional criteria for perfect reconstruction is not no longer
0:03:50if we want to have a perfect reconstruction would have to follow the this new criterion that is
0:03:59below little here
0:04:03okay that
0:04:05the point is that the blocks can not be a concern
0:04:08in that in
0:04:09for a construct this C no or
0:04:12a a does this know we are going to have to like late
0:04:16the contribution of several neighbour block
0:04:20and a that that i and not the difference from block to is that in the or the since we
0:04:25don't have all the name but
0:04:27neighbor block
0:04:28need that
0:04:29we have to care
0:04:31um a the tree
0:04:37see again uh
0:04:39the frame rate
0:04:41most so i i believe most of you is already
0:04:45familiar with this speech
0:04:47we well i wanted to remember that
0:04:50even though the it's not it's it's for what was proposed to be a video coding standard
0:04:56when used to encode
0:04:58in my
0:04:59you in my just
0:05:00it is a very efficient a image coding
0:05:04in one of the key features
0:05:06why i
0:05:07happens used the
0:05:09implementation of
0:05:10inter but each
0:05:12that consists on using the immediate neighbours
0:05:16but some neighbours of
0:05:17because of the encoder there close to that
0:05:20to predict the a lot were and coding at this moment
0:05:24and in this way you just have to send the with
0:05:29here you can see that nine and prediction modes
0:05:31for a four by four in a by lot
0:05:35and we have additional
0:05:37for a prediction modes for a lots of sixteen by sixteen B
0:05:43oh Y
0:05:44what's the dish you could of using this is scheme in a lap but transforms
0:05:50and uh the for the of
0:05:52because to a good is
0:05:54C we it's than in it was at local
0:05:57not the words would have to have the a construction of the peaks as
0:06:00for use
0:06:01it in the prediction
0:06:04and we as we so
0:06:06we don't have the have construction before having all the neighbours
0:06:10we can not perform the
0:06:13the prediction of
0:06:14a a a a a a a a a broad we are we want to go be
0:06:19the colours
0:06:21but uh uh the central out here the
0:06:23being the block we are
0:06:25going to in
0:06:27using a that are solved we would need the
0:06:31and square
0:06:32but we have only this region here
0:06:35that a use is available
0:06:37for a constant
0:06:40we propose that instead of predicting the block itself that like to pretty the this is standard block
0:06:46in this way
0:06:48every single pixel so outside the bar
0:06:50far away from the as will be predicted for time
0:06:54for our overlapping factor of two
0:06:57yeah yeah the if we want to do this
0:07:00we would do is to be me sing
0:07:03half of the peaks as in the last one
0:07:07and the black you hear have have not not a been
0:07:16if should but we go to the way that
0:07:18jpeg that this
0:07:19it's not a six four in the block in which
0:07:23a about is
0:07:25a true putting a a a larger block
0:07:27we have a block of it by it is that the but in a
0:07:30model a block of sixty but sixteen
0:07:32and the blocks so oh oh O do in this way
0:07:36we can see that for the first block
0:07:39since the for a lot of the proof isn't mark or bob had already been called
0:07:44we need this case would have all the peaks is of variable for the prediction
0:07:49however when we go to the set and more
0:07:52we would go back to the the case we had before where
0:07:57the last
0:07:58neighbours are not of a
0:08:01in the
0:08:03becomes worse because we have all rows
0:08:07half of the pixels in arts and uh
0:08:10and in the four we would only have
0:08:14i was more or not
0:08:16i don't know
0:08:18see i D is is more corner to predict
0:08:23the block
0:08:25to prove a little what we probe one of the things we propose is changing the order of encoding that
0:08:31this plots
0:08:32changing the had of those set and then the block
0:08:36in this way in the first we have the same conditions we had before
0:08:45in the set and we go to the same question all
0:08:48the same case we had in the fourth level
0:08:51which means that uh
0:08:54we have just one corner in this case it got worse
0:08:58however i that are that we do didn't have the first the the left corner we again have so
0:09:05in this case we are going to ensure sure that half of the box will have all the pixels are
0:09:10for the prediction
0:09:12and and the that have
0:09:13we have just this corner
0:09:15in this case we are going to propose to use just a D C
0:09:19a prediction that will consist of a pretty can all of the pieces
0:09:24as the average of the available peaks
0:09:29the process will want to you know
0:09:30a a much seem has done in in a you know the
0:09:35the the residual be couplet is difference
0:09:39that that he's a big so and the prediction
0:09:44after the
0:09:46quantisation is a and inverse transform
0:09:48i can we have to remember that the it is you do you only if we did not have the
0:09:54is different from
0:09:56the original one
0:09:58and scenes
0:09:59we we have to mimic a perform adding the prediction what have to mimic eight
0:10:03this process
0:10:05in the prediction
0:10:06we simply we simply process it
0:10:09the prediction of before we need to the was you to to obtain that you have a we with the
0:10:15if we didn't have the prediction
0:10:19also also have tool
0:10:21change though the way that
0:10:23the mode so it's selected
0:10:25in we have maybe two
0:10:28mm make two wave of
0:10:30so like like no one is the reduction of
0:10:33the minimization of the prediction
0:10:36that's normally in measured by a sum of of bits a the difference
0:10:40which doesn't seem to be up but in this case because
0:10:45a is a big part as and the have construction of every C
0:10:49pixel so will be different depending on the position
0:10:53so we propose to weight the i and that this difference
0:10:58a going to need to the importance of the
0:11:01of of you of its simple
0:11:04uh and the proposed a weighting in is that we use in this work is
0:11:09given by these that would be the have constructed
0:11:15have that we would have if we had that much it just with one
0:11:21the rate distortion optimization can also be performed a but we have to remember that since we cannot not construct
0:11:27the plot
0:11:29the distortion has to be a and and measure in the transform domain and the a for the norm
0:11:35orthogonal transform
0:11:37we have to take a called the energy of a
0:11:43this is the uh one example of weighting matrix
0:11:46or the you generalized lab
0:11:48of the hot transform that at but
0:11:51well the few as well or thing by maximizing the coding gain for uh out of a grass is model
0:12:00correlation factor of zero point nine five which is
0:12:03not to be a a good model for in
0:12:09and now we go to
0:12:10a and we will present uh
0:12:15a so of for and the implementation that idea at at
0:12:18this point would was just to prove the concept that that
0:12:22the the of sounds can be used the again at
0:12:27that but does ones could use a with friends good frame prediction
0:12:31so for simplicity reasons
0:12:33the than a dish was scaled denoted by we'll
0:12:37and because of this court the data
0:12:40coding but that was six at two eight by
0:12:44also we use a the general like to be a talk or not a song that would result
0:12:49for several reasons uh
0:12:51but the overlap you of overlapping factor of two
0:12:54maximise the number of lots that we can use the pretty
0:12:59the she she B T has a a good performance for this overlap factor
0:13:05and uh uh uh just every since the overlapping factor or or was of to
0:13:10we had a prediction block that once of sixteen by sixteen
0:13:15and different or we can implement any number of prediction modes
0:13:19at this more that's point only the four modes of a able in the H that six four
0:13:24were implement implemented
0:13:26when we are going to use
0:13:27let and so
0:13:31for comparison reasons
0:13:34also the traditional intra-frame prediction was implemented when using uh dct
0:13:41in this case all the nine modes
0:13:43were available
0:13:47including them
0:13:48the mode
0:13:49we use we we is that the same approach to
0:13:52a propose for it's that the six for that
0:13:55means that means
0:13:56looking at the
0:13:58that an upper neighbour
0:13:59we but they the best uh
0:14:02the most probable mode
0:14:04and we can that in
0:14:07or the proposed of this scheme since the a or a neighbour
0:14:12didn't have that
0:14:13several models of prediction
0:14:16was seen play just to use to bit cold
0:14:19to to the half of the blocks in for the order have we don't have to encode anything "'cause" there
0:14:25is just one
0:14:29besides that that that application
0:14:32which means
0:14:33no additional like
0:14:35of flat but that's forms were also that
0:14:41you can see for uh
0:14:44are they
0:14:45imagine a better
0:14:46we can see that uh
0:14:48the proposed
0:14:50method that uses in blue
0:14:53are performed
0:14:54the transforms
0:14:56and then you a plus dct that we plan
0:15:00we can see also that most of taking in this case was that just by simple education of love but
0:15:07or some but but again
0:15:08our proposed method the
0:15:10uh use
0:15:12for the
0:15:12improvement in the
0:15:17this same result
0:15:18signal as a
0:15:21by the first frame of the for which D C can
0:15:25give a bad
0:15:26in which again way our
0:15:29uh propose a method the or performed both in
0:15:32you inter a C T and
0:15:38and then can and we have for several other image
0:15:41in we she we can we can see you the result
0:15:44for the you the but not much
0:15:48or more comparing to
0:15:50in to plus D C T and and the right to
0:15:54we have uh lap buttons so
0:15:56we can see if here that we have a no
0:16:05now we go to the conclusions
0:16:07the results presented here
0:16:09sure that the entrance to addition could be adapted to be compatible with like
0:16:17also also
0:16:18we show that this
0:16:20propose a scheme are performs the application of lapped transform as well as the inter prediction with this at the
0:16:28you all tested in
0:16:31important to note that in our case we have just half of the block
0:16:36but the being predict and the
0:16:39new in have we have only
0:16:42prediction modes we so if we implement that all the i'm not a different number of the
0:16:48results presented you
0:16:50could be for to improve
0:16:54we have a also it's a very preliminary
0:17:00this is going to imitation of the
0:17:03of the scheme presented the here in a a real which
0:17:06that's six for older
0:17:08we can see here that even though the gains
0:17:12i smaller
0:17:13we have a
0:17:15in all tested
0:17:18for the image by
0:17:24a few
0:17:25reasons why the gains are a smaller it it's that in this case where competing with
0:17:31not all
0:17:32not only the nine modes
0:17:34of fate by eight but
0:17:35then nine modes of
0:17:37four by four and the for of sixteen by sixteen
0:17:40and in our case we have only
0:17:43them for modes of
0:17:46that time limit in in eight by
0:17:50uh coding block
0:17:51also the one
0:17:53and a but may not be as well as a a a at that that
0:17:58compared up to compress in
0:18:00a a lot but transforms transform utterance from
0:18:05as is
0:18:06in the dct
0:18:10i four
0:18:11sure where we are what what i've seen
0:18:15i'll way of
0:18:16implementing infallible
0:18:18size prediction block
0:18:20as well what that thing out of a lot of factors
0:18:23to see
0:18:24we use what we lose in the prediction we can gain in the near future
0:18:30and and uh um we all i want to say you also that a station of this work has been
0:18:35a set than in i C two thousand and
0:18:38they extension to feed you
0:18:41sept and will be presented nice
0:18:45that can
0:18:47presentation thank you and
0:18:57any questions from now
0:19:04i one quick question can you comment a computational complexity group
0:19:10okay just
0:19:11thing is the the suit uh
0:19:13in this case for times B
0:19:15the prediction will be done
0:19:17for time you can even though i did manage that
0:19:21that's see
0:19:22i can say that will be at least for time
0:19:28a the in the part of prediction hand
0:19:31in first and
0:19:32the direct
0:19:36but the point point
0:19:37the that or the power will be seen it to the
0:19:40it's that
0:19:49but was of speaker