so um my name is a somewhat what an bit from japan and so today i would direct to talk about uh about my a a a little work uh tight so that all fast identification of jpeg two thousand images for distance there's to provide uh this is a i'll try all of my talk part a i'm direct talk about C but grand all of this work a second yeah jpeg two thousand coding system and it's coding the power as uh some rice then then the calm an image identification miss out of what chip pick two so than images is introduced and in the program all was calm based a miss not uh describe the procedure zero a missile to comes next for the by experiment in results and if yeah i it's at time i we conclude my what um did does him uh is no no as one oh uh the most successful up of case one all say pick two thousand uh jpeg peg two thousand has a coding provides for just as cinema application as an i mean doesn't and all that to standard uh in this provide a uh coding parameters uh all typical two thousand made if a for it's provide um in this just name and applications uh such as a and being or we in and so on uh there are needs to i didn't to find and exact frame exact the same frame which has the same or is no image as a given or is not image uh this identification called image identification and uh this identification of jpeg two sounds and images are required a it's its S name because uh frames all this that's the contents uh encoded by a typical two sets and so um image i didn't you meet identification is defined uh an operation for finding an image identical to a a given the image from you may database and that this processing should be done in compressed domain uh how about uh commissioner miss thoughts cannot handle the chip to send images support just them um because in this this smell provides uh coding parameters such as a the average a label on and that quantization step size and the call broke side or or and so on um may differ um now i would like to talk about jpeg two thousand coding system this is a a diagram or jpeg two thousand coding system and uh decoding power meters using and jpeg two says uh used in jpeg two thousand uh according to each operation broke uh a was that rate character so uh i fast uh input image is uh and i Z by D that would be and think ties it and they're them uh each transform questions uh divided uh called rock and then bit about it meet the ring a uh it's down so after a bit modeling the number of bit of brains uh it called in this presentation a a and and that B P uh is produced and the recorded it as a part of the had that information of jpeg two sounds and called the story uh in this work uh this and that the bp is used to fall image identification uh i would like to talk uh that that detail four uh and that P P uh this figure shows uh transformed questions uh this is a wherever it questions and the to this it's a seven one and uh the rocks uh here so called rocks uh each quantise questions is separated into this each sign and the i was sort it should and the where a we have the absolute to use of for tries a speech prince from the most significant bit to say the least significant country all the samples in the bit prince uh is out zero or one then is a special breed brain uh in fish all samples are zero it's a P S and this special bit-plane is if to as a zero a brain and then when all it prints out zero be brains uh the cold broke this uh this is i coding you need to a chip pick two cells and uh defined as a not included broke because it does not include any data that to be encoded uh this figure shows the example all the and P P and not that uh and that B B can be easily extracted by passing to hit that all jpeg two thousand called streams and uh no know coding is it and uh compression late or was it called a stream has no impact to to bad in that P P uh there is a metal two vol you mean image identification using and that the P P uh i would like to summarise uh miss a uh this ought passed to extract the and that to be P using a hit the se and then uh the components and all in to be P between a query eight big two thousand images and the i image if mean and the J peg two that the images in uh and you made database uh this mess of works where for the a you is that to have the same coding power as however about it is not applicable with different coding parameters because uh barry all and that the P P in images compressed to is different coding parameters may be different so two so was uh this program uh the extended ms of it's probable uh the proposed missile a is composed of four steps at fast uh regions to be compared to a defined using the regions uh the difference of the labels and the or code called broke size can be can say uh here is a transform the sent uh of a query image and a this is the database one and uh each a a query images uh be composed to by W dwt and level four and don't data bases a three and we choose uh many little between the images uh for for identification so this is just have a one so in could we made this uh the rocks size is set to a to by set to but uh in database images uh that the uh for group size uh it's set to sixty four by sixty four uh they are define in each other the uh the whole so is on top and the but can sides off compared regions uh sit to user recent marks marks spot or this site so uh this is a example compared regions uh one regions uh has a a a a full whole blocks for a query images but uh one regions has a a one of the broke for database minute uh if but blue number of did zero bit to in the region like a query images uh or is the minimum value is used in some so you can't process like is uh a few days not included but you uh minimum value is you priest the by not good and then uh omission mission or relation vectors uh a right so the and that B P all joe for a region this is our region a region and that it's okay neighbouring regions are can be a and then so a back that is a produce this value all one means a in that B P or was a neighbouring regions is small runs and for or one and the minus one means a a and that to be also a neighbouring regions is run and folk or one and uh the components comparison with anybody value and or between uh not to include it oh to put zero eight minute all was a big to oh finally the local decision uh is made by these operation this local decision it's four one compared regions and the when uh for or uh this is a is or all or all or can the regions uh but steve this shows are made so the global but this is a is may if all local decisions uh point steve and then colour global but this isn't is plus still in the experiment uh we use the standard the evaluation just marked area uh it is defined as a uh D C i'm uh this never and you said tape uh was used uh this just it contained a uh over fifteen thousand frames and uh have fast a sequence of jpeg two thousand compressed you may is was to in a database and uh a query images once found a and uh and in five identified to an the jpeg two thousand images in that data uh the identification experiment how was performed to for all possible combinations of a query and a database images and the number of combinations of a query and the database well as a fifteen so was and by fifty and in total yeah it's a property of of the test C against and then the encoding power we can support query images uh the resolution all this C case is a four sounds and by two thousand and uh this is a so as as a coding a permit to as so and the this table a a is the power as support database images in this table uh this did to you piece if you've five three and C B C six people C B want to eight are we is i tape uh means uh images encoded using a different do you double D V you as and the core the block is a divide to iteration and different step sizes uh this is an experimental results and the the is force was you right if B R and the two to rate T B R is uh kinda kinda correct by these equations this table shows so the idea of identification based on the proposed a of produce to and uh one percent all the a a one percent if B are we of the difference of jpeg two thousand coding probably and the result any false negative match and uh this experiment a X Q to uh on a workstation with uh wrecked is and that the is processing time for one combination of a curry and uh database a image was about uh the point six me second pass frame a a a a finite i really i would to conclude the my talk uh and identification method of what jpeg two thousand images with uh different jpeg two thousand coding parameters has be present it uh for each as uh the proposed wasn't miss out is uh compressive domain processing and uh has a robust S against a difference encoding calling S and uh it's one runs last thank you for your attention any questions i i i one push i you should in the previous slide that but the uh i it T P R of the true positive and it was at a hundred percent does it mean that you could actually get results that the overall better but proving the F P R introducing reducing P are a little bit if you change the threshold a a actually a a we we have a we have a for home to some experiments uh a with a a different coding rate so we get the very low rate uh some results a uh oh obtained uh not hundred in "'kay" but uh oops also so much of different a a and you the questions Q let's the to speaker