290 likes | 503 Views
Final Project : Motion Estimation. Adviser :葉家宏 老師 Presenter :吳奕賢 Date : 2011/01/13(Thur.). Overview. Background Sequence analysis Proposed search Method Demonstration Comparison Conclusion Discussion. Background.
E N D
Final Project:Motion Estimation Adviser :葉家宏 老師 Presenter:吳奕賢 Date :2011/01/13(Thur.)
Overview • Background • Sequence analysis • Proposed search • Method • Demonstration • Comparison • Conclusion • Discussion
Background • Implement motion estimation on Akiyo, Football andForeman QCIF sequence • Three different basic techniques: • Full search • Three step search • Diamond search • Coding time and PSNR comparison • Microsoft Visual Studio 2008 C++
Sequence analysis • Akiyo
Sequence analysis • Football
Sequence analysis • Foreman
Proposed Search • Step 1: Pick an initial step size. Look at the block at the middle search, the four points are at a distance of four pixel from center on the X and Y axis
Proposed Search If minimum SAD point would at center • Step 2: Diamond search-point configuration
Proposed Search • Step 3: Next step along a diamond’s vertex
Proposed Search • Step 4: Final step with a shrunk diamond
Proposed Search If minimum SAD point wouldn’t at center • Step 2: Square search-point, which are at a distance four from near point
Proposed Search • Step 3: Diamond search-point configuration
Proposed Search • Step 4: Next step along a diamond’s vertex
Proposed Search • Step 5: Final step with a shrunk diamond
Demonstration • 1st sequence : Akiyo.qcif
Demonstration • 2nd sequence : Football.qcif
Demonstration • 3rd sequence : foreman.qcif
Experimental Results Comparison
Experimental Results Comparison
Conclusion • If we only use Three step search, it couldn’t find the exact motion vector than Diamond search • If we only use Diamond search, it might be unable to search the marginal point of search window • We combine the advantage of the Three step search and Diamond search • After sequence analysis, we improve the method at step 1 that could reduce the ME times • Focus on PSNR, The proposed search is better than Three step search and Diamond Search
Discussion • Step 1: Square search-point configuration • Step 2: Next step along a square’s vertex • Step 3: Final step with a shrunk diamond
Discussion • Step 1: Hexagon search-point configuration • Step 2: Next step along a hexagon’s vertex • Step 3: Final step with a shrunk square
Discussion • Step 1: Cross search-point configuration • Step 2: Next step along a cross’vertex • Step 3: Final step with a shrunk square