0:00:13a
0:00:14the and they have an uh i'm and
0:00:16a uh i'm from the that's going from a the problem
0:00:19uh and you estimated quite spar
0:00:22and that type on map and bases
0:00:24is a reason against sport image you the creation
0:00:27and this voice collaborate with uh
0:00:30yeah and got and the for a radio
0:00:33so uh
0:00:34yes the a around this print is and uh for our uh uh uh introduced the brief can set up
0:00:39a you listen again
0:00:41and then and i have a
0:00:42at discussed how to formulate the image into to and you is again
0:00:47and by are so some in is spends a read all for or by some conclusions
0:00:53so uh seventy again
0:00:55maybe you C components
0:00:57and the first compress prayer who prayed again
0:01:01and here is by note the base you one up to you and we assume they are and prayers S
0:01:06and this and compare use the use straight is set
0:01:09and and here
0:01:10a a a i ten for the few straight is set for use at you Y
0:01:15and the
0:01:16that that i uh i think so as a at a i i is the
0:01:21all of possible uh a Q strategy that that use so you like can pay
0:01:27and this can put a is the feel function
0:01:30we is uh with determine the pale that that use so you uh a that use the can opt ten
0:01:35by taking a something and
0:01:37a so or the add i stand for the uh
0:01:41little pale for a single of you uh you i
0:01:45and with a note is that uh all the at i
0:01:48user is not only to turn you by le use you ice X
0:01:52but also all allows users X
0:01:55and and yeah i us some a to do this and that we use in this print
0:01:59first is the strategy profile
0:02:02a
0:02:03is that class less and all all the players X and you want to a a and
0:02:07and this strategies is it a file is set pretty a
0:02:11a you i use a neck i
0:02:13uh so is the and it again the class and of the all operators
0:02:17uh actually it's seven player you wise X
0:02:21and the P say set as i say is that content all of possible pure strategy let the use that
0:02:27you i can take
0:02:28and that means strategy G is a random might a strategy
0:02:32uh well
0:02:33can be
0:02:34represents as pi
0:02:36we is a probability distribution of these
0:02:39Q trace set
0:02:42so uh
0:02:44to mess my on the apple and pale a prayer and it's know all information all are out of videos
0:02:51uh and channel is very difficult or even impossible to for prayer that's and no out a real a strategy
0:02:57and P
0:02:59so i that's K
0:03:00to improve the P L
0:03:02a a a a uh we are let hard to predict the N using the um um target to you
0:03:07uh
0:03:07oh understanding by then uh by building
0:03:11and this this to the concept you loose in again
0:03:14so and people is again is that and that's not be the interest and dependent strategy in population
0:03:21uh and uh a was set in that you recent again is slightly bruce and they stable strategy
0:03:26which is
0:03:27but strategy is that's that if all band bounds of the part this um top it
0:03:32then no me intense that you can invade the populations and that in inference
0:03:37well mistress and
0:03:39and such and
0:03:40and you produce an is they was to you can be simply and using the updating function here
0:03:46and
0:03:46uh the idea uh behind is updating dating thing is that
0:03:50uh the probability of using us that ten
0:03:53strategy
0:03:54at time T plus one
0:03:56is it to me is proposed a note to look pay all
0:04:00oh using using this strategy a previous time
0:04:04T
0:04:06so a with the background of the you risen again now you're like uh should discuss how to uh for
0:04:12me the image into and as a duration again
0:04:15so i think you in their present problem we have the
0:04:19uh a less as a image uh i here the
0:04:23but so but i know the low less was and not that is
0:04:26two
0:04:27can a high does to the image
0:04:30and and in it is a rate are they ask
0:04:32many but um is that mean proposed for this problem
0:04:35and here we only consider the uh was best so
0:04:38but that the if you're attending
0:04:40well for it's P so i had a it's uh for it uh it's be sides i see
0:04:46uh we use uh
0:04:47a in a combination of a set of uh a low less and
0:04:51this so i out of here
0:04:54i use is we assume L uh ways and non negative and this sum up to one
0:04:59and and if we know the this that a the low less and P so yeah the problem not be
0:05:05how to find a way
0:05:07so uh or is that we know that the team like can not be of then
0:05:11because um
0:05:13we do not have a or in a high uh less to see me image
0:05:16uh and a look up previous approach
0:05:20i i estimate the weights
0:05:21based on the sounds and that the
0:05:23could balance makes use of a high does a
0:05:26you means it can be where estimate form no combat about its you of low low say image
0:05:31i one is the sounds a not true then the in to present a performance
0:05:36we have be great thirty degraded
0:05:38so he a
0:05:40is that a a in the way in one that
0:05:42we propose to progress the we find the way by at an antibody that the made in the ways based
0:05:49on consistent
0:05:50and then finding the look and and using the ways
0:05:53as as a refinement process as is at sri and you produce in a process
0:05:57so which can be naturally really one it S and you resume again
0:06:02so uh
0:06:03yeah is the uh units there for me i say it's very simple
0:06:07we you see that a a in this uh a one is and the prayer a to a no high
0:06:11does of them so
0:06:12i it's i G
0:06:14and the use i T are the corresponding little as an piece so i have here are
0:06:19and that use as that is that can do this ad
0:06:23and that me is that the uh the outcome of that means strategy is the estimate of the high does
0:06:28and piece shows i had it's
0:06:32and the probability in the this strategy can be do guys the negative
0:06:37uh norm rice a topic you
0:06:40so we so kind the formulation
0:06:42uh the problem of in the a
0:06:45S the mail to find a an piece so we can the problem of finding the people produce a stable
0:06:49strategy of this again
0:06:53so a a how the size now look Q said to hear a uh we use use a very simple
0:06:58up used strategy said
0:06:59well for it's a a high to and be so we just use the the
0:07:03uh
0:07:04a square window of by than the is low less system this shows
0:07:08example here uh uh we have the
0:07:11i sh uh this uh this we uh a hard as and P is uh i
0:07:15two and my as one to and
0:07:17and i two and two and minus one and i two and to N
0:07:22and with just use the scroll window as and about a
0:07:25the P so look case that's a as one and that's one minus one
0:07:31and
0:07:32for a pale constant and a
0:07:35we we force uh define the
0:07:39similarity measure between uh can it is uh
0:07:43twenty it P so the i C and the time P so i it's i G
0:07:47and these uh uh uh
0:07:49similar he's so that's by lot you could eat it this and of the
0:07:54oh local pet sent the by the i'd G
0:07:56which is be yeah i G
0:07:59and that of the local patch since the by i it's i'd say we is to be i side G
0:08:05and from this uh a a a it's of the and we can see that if a i'd T close
0:08:09to the i as i'd say then that source and we're be
0:08:12um a small
0:08:14is that's K with to give a higher payoff O
0:08:17a a high P L to that uh can it be so
0:08:20and only out a hand if a i'd say is far away from i i say
0:08:25then this just all since to be lots
0:08:27and use that's K we give a low pay to the
0:08:31look can the piece source
0:08:33so is all sets uh in you uh in deviation a lapel since to be a six creasing function of
0:08:38this D yeah G and are we use the as one in seconds constant
0:08:44and
0:08:45was a know this so that for uh
0:08:47can it is to look at a hard put used an uh which actually use the we and just a
0:08:53be at times no uh time and S
0:08:55i had a it's T I G
0:08:58and if we compute a you could it this than you would be zero
0:09:01but that's to the beacons that's some may have some that also
0:09:05we cannot use that um as a matter
0:09:08so he we we use uh a very simple is that
0:09:11just we is that this also using the minimal distortion or or or a duck is so
0:09:17and again let P L forms an all is used to compute by a a once and
0:09:24so uh
0:09:27so that's look at some uh i spend the out so we test all mess so
0:09:31a to you need to did not board
0:09:33and we assume that lower low those an image is is down sampled from a high does and you image
0:09:38and we assume or is an image E
0:09:42is a battle
0:09:44for a a a as a been marked
0:09:46and we compare with the by Q P and S A are misses
0:09:51so if here is uh the result for a D a V G a
0:09:55the A the low lows the image can be is the result of open by by Q P
0:10:00and C is an a result uh generated by a the i and D is the
0:10:05read all all misses
0:10:07we can see that uh
0:10:10uh B C uh the are some a big at fair here
0:10:14uh we saw mess so we can a voice that's kinds are but
0:10:17maybe can the good is more clear
0:10:20uh for a boat image
0:10:22we can see that they are some fig for that
0:10:26we we a mass so we can uh a great the uh read you this kinds are the pen but
0:10:30you'd better read
0:10:32Y thing which T is that is um
0:10:35in the original boat image yeah some since noise
0:10:40and
0:10:41in that case if uh you
0:10:43use a a a i to be and it always yeah i to separate such kind of image the noise
0:10:48would be puts i'll and this at to use re though closing the figure that out that
0:10:54and
0:10:55you we saw mess so we we can as we can suppress such kinds of sense noise
0:10:59which being used as a a mess that have the potential to perform joint in to and
0:11:05and and this is a body if i to using this is but to here
0:11:09a a we assume uh
0:11:11less less as say just noise
0:11:13it's it's to buy some calls and noise
0:11:16uh and we compare our our that we sit them as a lap was performed a denoising using the long
0:11:22local and then performed in to place and using
0:11:25uh by Q be all the
0:11:27as
0:11:28we can see that the S some are right
0:11:30you
0:11:31well the phase and the no it's region
0:11:34and we
0:11:34and this kind of a a a page uh maybe come um many from
0:11:40uh this separation of that the anointing a to prison
0:11:44uh
0:11:45which means that you you you or did know any i wouldn't can not if we moves noise
0:11:50then the noise would be pushed are you in in the in the present process
0:11:54i we saw a mass so uh we can
0:11:56performs time it ten is at the non you can to raise and and we such kinds of sewing a
0:12:01process we can suppress such kinds of the play
0:12:05so um
0:12:07a come to can goes and uh so we propose a
0:12:11but are from form actually a general framework or a image uh
0:12:15in the present and using that you person again
0:12:18and this kind of for what it's very simple
0:12:20and and in a few that we like to investigate hall fall out uh that we can uh and uh
0:12:26good P L problems and to improve the result because now a
0:12:29the P L the as is very straight for just using the it read it
0:12:34this than uh between the low local patch
0:12:37and we also uh
0:12:39uh we also in in
0:12:42a a gem a to investigate the possibility of gen lighting but addition approach and all framework
0:12:49and just can my
0:12:51vol
0:12:51the Q
0:12:57right so
0:12:58we we can't to thanks okay
0:13:00ah
0:13:01we can afford
0:13:03quite a few number of questions
0:13:09so
0:13:10um and you go back to the land uh uh uh in its uh
0:13:13yeah O
0:13:14maybe a couple more so what's the T P M
0:13:17psnr
0:13:18i i i i so i i segments that this uh a eyes only for this three all i see
0:13:22okay because i have never or or right before
0:13:26uh
0:13:26oh typically for the entire image are right five thirty six right right of L a little right okay
0:13:32yeah he even for the right keep but as i for a whole image is no
0:13:36it's key problems in that
0:13:39he a very naive questions um
0:13:42typically one we play games we compete
0:13:45uh_huh and and the play was trying to compete right
0:13:48uh
0:13:49but he you are posing
0:13:52the uh be interpolated pixels is play the right and i don't see why
0:13:58okay okay uh as the competition so why to why needed so for the given to the are two kinds
0:14:03of the many yeah why as long at which the uh pretty a compute which is a i'll a a
0:14:09and one type of a not type is the culprit again
0:14:12well i
0:14:13pay of pretty which is a lower so
0:14:15for for these will lose it again this the corporate again
0:14:18so they call in step
0:14:21compute you which is um
0:14:23so yeah so
0:14:25so basically what how they cooperate with these are is that uh
0:14:30for example uh we use this uh
0:14:34when we compute the
0:14:35P of phones and
0:14:37we were use a week can just and all
0:14:39this just so for this so
0:14:41a so in such case are if they have
0:14:44uh
0:14:45very high similarity then they we help each other to what you're better results
0:14:50yeah
0:14:53um
0:14:54and
0:14:55obviously is as
0:14:57it changes
0:14:58uh
0:14:59from from one pixel of the uh the other the weighting the resulting optimal weights that you get out at
0:15:05the end is
0:15:07is
0:15:07is uh as a function of each
0:15:10pixel corps
0:15:12not not by patch but by
0:15:15yeah just for it's piece that the for each voice of right yeah wait yeah each pixel right
0:15:20uh do they have
0:15:21i all have to for little exactly the um
0:15:25some up to one right you need to have some up to why how to wise you have
0:15:30different being the intensity with a deeper
0:15:33the was
0:15:33each
0:15:35any of the question
0:15:38okay if not uh
0:15:40hopefully a use right
0:15:43okay
0:15:44then