비디오 압축에 있어서 빠른 움직임추정을 위한 개선된 1비트 변환 기법

Abstract

Motion estimation being vital to digital video compression is an integral part of video coding standards like MPEG-1, MPEG-2, MPEG-4 and H.261, H.263, H.264/AVC. In any video codec the motion estimation module consumes the most processing time and computational resources. To accelerate this process is must to achieve real time video coding. The field of motion estimation has been researched since long targeting to achieve high speed, low complexity and high data rates. The One-Bit Transform is such a technique in which the lower bit-depth representations are used to accelerate the block matching and hence the motion estimation process. This thesis is focused on the approaches and methods that further improve the speed and quality performance of the One-Bit Transform. In this regard different techniques are proposed which use various combinational approaches of existing fast search algorithms with the One-Bit Transform. The first algorithm namely the binary multilevel successive elimination algorithm (BMSEA) uses the multilevel successive elimination on the binary planes of the One-Bit Transform. Then to further improve the performance of BMSEA the use of median vector is proposed to exclude more improbable search locations. Also the idea of combining the partial distortion elimination (PDE) with the BMSEA is presented to make the decision boundaries more tighter and eliminating unnecessary calculations. The simulation results show that all of the above combinations significantly improve the speed performance for the One-Bit Transform based motion estimation. Finally a technique to improve the quality of motion vectors obtained by the BMSEA is proposed by using a slightly different cost function. Simulations on different sequences show that there is an improvement in the PSNR and perceptual quality of the reconstructed image.