Abstract: This paper proposes a block-matching algorithm named Small Diamond-Linear Search(SDLS) using small diamond-linear search pattern. For stationary blocks, SDLS can find the Motion Vector(MV) by one small diamond search. For blocks with motion, SDLS gets a block distortion decent direction from the distribution of computed block distortions. By using a mixed small diamond-linear search pattern, it can reduce the number of search points and locate the MV rapidly. Through using the search center prediction, SDLS can further improve the search speed and quality. Experimental results show that the proposed algorithm can decrease the search points by more than 50%, compared with Diamond Search(DS), Cross-Diamond Search(CDS), etc., while maintaining similar PSNR of pictures.

Key words: motion estimation, block distortion descent, line shape search, prediction search center

摘要： 提出一种基于小菱形-线形搜索模板的块匹配算法。对于静止块，通过小菱形搜索一步到位。对于运动块，利用已计算的块误差的分布，得到块误差下降方向，并用小菱形-线形混合模板快速定位运动矢量，使搜索点数大幅减少。通过预测搜索中心，使速度和精度进一步优化。实验结果表明，在保持图像信噪比的基础上，搜索点数比菱形搜索法、十字-菱形搜索法等平均减少50%以上。

关键词: 运动估计, 块误差下降, 线形搜索, 预测搜索中心

CLC Number: 

• TP18