Research Of Video Coding Based On Geometric Constraints

The video coding standard H.264/AVC achieves high compression ratio for its adoption of various techniques like precise intra/inter predictions, integer transformation, and highly efficient entropy coding, which also leads to high computational complexity. There lie geometric constraints between neighboring frames in both some single view sequences with restricted motion and multiview sequences. These geometric constraints are valuable for reducing spatial and temporal redundancy.In this thesis, the application of geometric constraints to video coding is studied based on the analysis of several constraints between neighboring frames in video sequences.For single view video coding, a novel homography-based search (HBS) algorithm for block motion estimation is proposed based on the study of the homography constraint between blocks in the sequences captured by PT cameras. HBS is then applied to code PTZ video sequences combined with the feature points matching method. Experimental results show that compared with than traditional methods, the proposed algorithm could reduce the computational complexity greatly and save the encoding time with negligible changes in video quality.For multiview video coding, a new geometric constraint called the Loop-Epipolar constraint is first proposed. On the one hand, the new constraint could deal with some shortcomings when utilizing the loop constraint in block-based video coding. On the other hand, it extends the application of the epipolar constraint from disparity estimation to motion estimation. Based on the Loop-Epipolar constraint, a new motion estimation algorithm is proposed for multiview video coding. It has been proved that the proposed algorithm is suitable to different sequences under different multiview camera arrangements, and it has a better performance than some new fast motion estimation algorithms.