A Study on Multi-bit-resolution Motion Estimation Algorithms

• Main Authors: Yung-Shang Chen, 陳永上
• Other Authors: Ching-Huang Wei
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/92315277502225078787

碩士 === 國立高雄第一科技大學 === 電腦與通訊工程系 === 90 === In this thesis, we describe some multi-bit-resolution motion estimation algorithms. Our proposed multi-bit-resolution motion estimation algorithms first reduce the bit resolution of the pixel values and then exhaustively match all possible block in the search window of low bit-resolution image to obtain a candidate motion vector (CMV) set by using a simple block-matching criterion to reduce the computational complexity and hardware cost. Secondly, the motion vector is refined on the positions of the CMV set by using the sum of absolute difference (SAD) block-matching criterion in the full-bit-resolution image. In our simulation, three measurement criteria, which are the peak-signal-to-noise ratio (PSNR) in the predicted image fidelity criterion, the entropy of prediction residue image, and the computational complexity of algorithm, are used to appraise the performance and the efficiency of the block estimation algorithms. The conventional full search algorithm always approach optimal performance in the PSNR and the entropy measurement. We could observe that our motion estimation methods, except our method A, obtain better performance in average PSNR and entropy measurement over the LRQME method proposed by Lee, Kim, and Chae in 1998. Their performance is close to that of the full search algorithm. We also observe that the motion estimation methods using multi-bit-resolution search schemes outperform the spare search algorithms (three-step search, new three-step search, four-step search, cross-search, and block-based gradient descent search algorithms) in average PSNR and entropy performance. In the comparison of algorithm’s efficiency, we could observe that the computational complexity of the full search algorithm is so large compared to the most other motion estimation algorithms. The computational complexity of our motion estimation C-1 is higher than that of the LRQME algorithm. Our method A has less the computational complexity than that of the LRQME algorithm. The computational complexities of our method group B and method C-2 are close to that of the LRQME algorithm. Moreover, the spare search algorithms have concise schemes from the viewpoint of the computational complexity.