240 likes | 457 Views
Wednesday , December 29 th 2010. Final Project : Motion estimation. Arsalane Yasmine Bories Marie-Noëlle Métais Romain. Overview. Purpose Full search 1. Method 2. Demonstration Binary Search 1. Method 2. Demonstration Three step search 1. Method 2. Demonstration Comparison
E N D
Wednesday, December29th2010 Final Project :Motion estimation ArsalaneYasmine Bories Marie-Noëlle Métais Romain
Overview • Purpose • Full search 1. Method 2. Demonstration • Binary Search 1. Method 2. Demonstration • Three step search 1. Method 2. Demonstration • Comparison • Conclusion
Purpose • Implement motion estimation on Akiyo and Football QCIF sequence • Three different techniques : • Full search • Three step search • Binary search • Coding time and PSNR comparison • C/C++
Full Search • The current MB must be compared with 225 candidate MBs of the previous frame if the search region is ±7. • Since an MB is 16×16 pixel, there are 99 MBs in one frame for a 176×144 pixel video. Therefore, the SAD (the sum of absolute differences) is computed 22275 times. The SAD is defined as follows :
Full Search • The SAD is defined as follows :
DemoFull Search • 1stsequence : Akiyo.qcif
DemoFull Search • 2ndsequence : Football.qcif
ThreeStepSearch • Here is search window where now we only consider a set of blocks (and not every blocks of the search window as we did in Full Search process): • The algorithm was introduced by Koga et al in 1981. • It became very popular because of its simplicity and also robust and near optimal performance. • Coarse to fine search pattern.
ThreeStepSearch • Step 1: An initial step size is picked. Eight blocks at a distance of step size from the centre (around the centre block) are picked for comparison. • Step 2: The step size is halved. The centre is moved to the point with the minimum distortion. Then, back to step 1 until step size equals to one.
DemoThreeStepSearch • 1stsequence : Akiyo.qcif
DemoThreeStepSearch • 2ndsequence : Football.qcif
BinarySearch • Step 1 : The MAD (Mean of Absolute Difference) is evaluated on a grid of 9 pixels that include the centre, the four corners of the search window and four pels at the boundaries. • The search window is divided into regions based on these points. • Step 2: A full search is performed in the region corresponding to the point with the smallest MAD.
DemoBinarySearch • 1stsequence : Akiyo.qcif
DemoBinarySearch • 2ndsequence : Football.qcif
Coding time Experimentalresults • Akiyoqcifsequence
Experimentalresults • Football qcifsequence
Conclusion • BetterresultswithAkiyosequencethanwithFootbballsequence: • PNSR value ishigher for Akiyosequence • Motion estimation isbetterbecausethe sequencehas a greatstatic part. • If weconsider the Akiyosequence: • The Full searchmethodgivesthe best resultsbut itscoding time is not optimal. • Binary search provides a better coding time but result about PSNR is lower • Three step search is the best one because coding time is insignificant and result about PSNR is almost the same than the full search.
Wednesday, December29th2010 Thank you for your attention ArsalaneYasmine Bories Marie-Noëlle Métais Romain