Abstract: By studying the spatial-temporal correlation and the center-biased characteristics of motion vectors in image sequences, this paper improves traditional diamond search(DS) algorithm, designs an adaptive large diamond search pattern (ALDSP), and presents an adaptive diamond search motion estimation algorithm based on motion classification information of blocks. ALDSP adaptively chooses different search methods with motion classification information of blocks. For fast moved block, it predicts initial search point and uses different search methods according to blocks of different relative motion. Experiments prove that the algorithm can improve the search speed in contrast to motion vector field adaptive search technology(MVFAST), when video quality is better.

Key words: motion estimation, motion classification, initial search point prediction, adaptive large diamond

摘要： 基于对图像序列运动矢量时空相关性和中心偏置特性的研究，对传统的菱形算法进行了改进，设计了自适应大菱形搜索模板，提出了一种基于块运动类型的自适应菱形运动估计算法，对大运动块进行起始点预测，再划分相对运动类型，对不同运动类型的块自适应地采用不同的搜索策略。实验结果证明，图像质量较好时，该算法相比运动矢量场自适应搜索算法，加快了搜索速度，降低了计算复杂度。

关键词: 运动估计, 运动类型划分, 起点预测, 自适应大菱形

CLC Number: 

• TP391