0:00:17uh i think it is my pleasure to present my work here and then a
0:00:22nice young channel and um now working in france at C inside your uh the
0:00:27subject of my paper is kernel similarity is that active appearance models for face recognition
0:00:35uh first of all i'd beginning uh i want to use a few words um
0:00:41active appearance models which is the base of my work uh it is quite to
0:00:47the widely used to yin uh face recognition and the object tracking and sometimes for
0:00:54uh medical image processing uh they in sexual uh the in central idea of this
0:01:01matter is to uh build a model which contains pose the shape and texture information
0:01:08uh of the training data uh and then when we have a new face to
0:01:13recognise the model can generate a pen generates the
0:01:18appearance of the new phase of id three shows so uh but us using this
0:01:24matter that we can get the detail of the appearance of a new face and
0:01:30also the uh location of the landmarks on the ball boulder of the base and
0:01:36the dog runs uh so it is still quite powerful wow algorithm for sometimes for
0:01:44the uh space
0:01:47for the recognition of the face expressions except for uh and the this matter it
0:01:54has it's lame eight so one of the most the important to a problem is
0:02:01uh is that the quite sensitive to the illumination conditions uh that is so why
0:02:07we want to our improvement
0:02:11and in recent years there was quite a lot of researchers work on this problem
0:02:16the illumination condition and the uh there is some idea is so what are some
0:02:22of them the posted to at some action a some action no parametrise scene this
0:02:29model which is the red delay corresponding to the illumination and to decide he rates
0:02:36they depend it to this work by using this site here uh active appearance model
0:02:41is able to generate the face is in a quite dark the illumination or a
0:02:48very bright illumination but uh we know that the illumination that is not the worst
0:02:54thing for the illumination when the uh illumination come from one side of the face
0:02:59it to make that is half dark and half right and this is the more
0:03:04complicated thing and the this idea that in the work for this case and someone's
0:03:12supposed to uh apply a filter on the decoder for two can uh invariance condition
0:03:19uh for example uh for here a transformation all couple filters uh and this matter
0:03:29the also cost to lose some data use the information from the recognition images and
0:03:36there is a kind of quite a tradition idea is to use some other october
0:03:40no uh transformation instead of principal component and that is to extract the uh the
0:03:47most important variables uh in that database and the although it is quite old about
0:03:53so we believe a we find the and up operate the transformation uh it can
0:03:58work for this case and this is the initial idea of our work
0:04:06and in this page i want to presents the database aware working on it is
0:04:11called a cmu pose illumination and expression database of human faces and descent also in
0:04:19this data database is captured the in section and environment uh in the room there
0:04:25is several cameras are wrong that the volunteer and several flashes are around him and
0:04:31to make a different illumination conditions the flashes flash one by one to get the
0:04:39illumination and the here i give out some examples in this database uh for each
0:04:46person your uh us fourteen different pose of that is and for each post there
0:04:53is twenty different illumination conditions and we can see from these pictures some of the
0:04:58illumination it's quite complicated but complicated and heart to recognise
0:05:05one so we decided to use this database uh with data statistic and the light
0:05:10on the database uh here in this page the we show that histogram of the
0:05:16euclidean distance between each of actors uh of the uh each vectors O and so
0:05:24one is for here is the uh that terms of the shape and here that
0:05:29is the distribution of the texture vectors uh we can see that so for the
0:05:35shape vectors it is the close to a gaussian distribution and for that extra ones
0:05:40uh it is so quite beautiful passion distribution uh this result is quite interesting and
0:05:47the according to this result we decided to use the kernel to be able to
0:05:53a similarity
0:05:57uh similarity matrix the instead of the covariance matrix which is to use the in
0:06:03pca uh have to see that it's uh occur no it's not such a new
0:06:10id reading in this case requires the kernel pca came out to maybe twenty years
0:06:16ago and it is directly used to eating active shape models which is the a
0:06:22priori is work of active appearance model but that the alternatives and to continue to
0:06:28use it in active appearance model requires it is very complicated to uh reconstruct the
0:06:35phase and we construct that you major from the uh extraction features
0:06:42but uh for the active appearance model it is very important to reconstruct the images
0:06:48um here the proposed method we call it to a kernel similarity component analyse it
0:06:54is quite different from the kernel pca but uh sometimes it seems quite similar with
0:07:01each other and to mathematics star race is quite so it's clearly reading you might
0:07:08paper he i don't want to uh talk about the can uh mathematical conclusions uh
0:07:14just the procedure of this matter what uh is very simple which is to use
0:07:20the kernel oh
0:07:22we just use the kernel to build the uh similarity metrics and then calculates a
0:07:27the principal component to from the uh
0:07:31from the uh from up from the metrics it uh and then we get the
0:07:36eigen faces which uh which represents a to the most important variation in the database
0:07:45uh and here in this in this page what i want to this i want
0:07:50to show how the eigen faces a facts to the variation of the model on
0:07:58the left part is the uh result of the proposed the matter uh we can
0:08:04see that for the first and the for the third feature uh it is obviously
0:08:10control the illumination variance uh the illumination environments on the face but so for the
0:08:17principal component to analyze um
0:08:21the bar uh the variance is only between the genders are sometimes between the different
0:08:28to uh shape of the face
0:08:32so this result is tell us that so we have already choose the appropriate the
0:08:41transformation um and here is the experimental results on the right C is the a
0:08:48bit evaluation curves which what i don't like it's uh i like to see directly
0:08:54the
0:08:56the image is as we said before for uh yes the first column is the
0:09:01result of the proposed method and the in the middle column eight is the result
0:09:06of standard again and left column is the original you make use a which is
0:09:11a to recognize and a as we said uh standard and it works well when
0:09:20the illumination is not that complicated but so when it is how dark and half
0:09:26right is uh the a and that and work but so the proposed method it
0:09:31gives the quite good result
0:09:34um we also applied this method in the a rotation of the phase in this
0:09:41problem but the uh from the U matrix we can see that the improvement is
0:09:47not that of years uh only for some certain case it's some uh some change
0:09:54but not a lot
0:10:01uh and here is the conclusion the proposed kernel similarity is the active appearance models
0:10:06is robust to illumination and pose changes of the pc images with this signal them
0:10:12at their depict the fitting procedure can accurately thing sizes bases for my right to
0:10:18my dark affected by the illumination and say i have to emphasize that this method
0:10:25has the quite big limit that sits requests uh applied set accuracy at a alignment
0:10:32of the shape and texture vector if we couldn't do this it won't work would
0:10:37and so that the next step for our work is that so we want to
0:10:41make the matter to work on both the pose variation and illumination variations
0:10:48and that this all thank you
0:11:07estimate the parameter of the
0:11:20so i can understand you clustering
0:11:25yeah uh_huh
0:11:30is your rights yeah there is a
0:11:34i you mean you mean i think you mean this one yeah this one oh
0:11:39yeah this is also part of our work
0:11:42you see wow build a histogram of the uh of the mecc vectors and so
0:11:49we uh and to that sir it shows the gaussian distribution so we just use
0:11:54the uh
0:11:57variance here
0:12:00it is a portion and to the kernel we use is the portion so
0:12:07we think here that it that is it's represents the variance of the gaussian distribution
0:12:15so is that the
0:13:28uh excuse me that's what is the other uh algorithm called you see
0:13:39uh yes i think i heard that before
0:13:58yeah actually the principal common to analyze is quite close to the uh independence the
0:14:04common is except that that's a we try this matter what right do you depend
0:14:09and one that's it doesn't work quite good
0:14:29oh yeah
0:14:35uh_huh
0:14:37yeah actually what have uh convert opens up the uh ninety eight percent each of
0:14:44the information