基于低频子图的运动估计算法  

Motion estimation based on low frequency sub-image

作　　者：余志雄[1] 孙洪[1] 贺军[1] 

机构地区：[1]武汉大学电子信息学院,湖北武汉430079

出　　处：《计算机应用》2006年第5期1024-1026,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60372057)

摘　　要：为了减少运动估计的计算复杂度并提高其搜索性能,提出了一种基于低频子图的运动估计算法。该算法首先将当前帧和参考帧通过低通滤波器得到数据量减为1/4的低频子图,然后用全搜索得到最佳匹配的低频子块,最后对低频子块所覆盖的原参考帧区域进行精细搜索得到最优点。实验结果表明,与传统运动估计算法相比,该算法提高了运动估计的精确度和降低了计算复杂度,并且对各类视频都有很高的鲁棒性。In order to reduce the computational complexity of motion estimation and improve the accuracy, a fast motion estimation algorithm based on low frequency sub-image was proposed. The algorithm firstly puts Current Frame and Reference Frame through a low-pass filter to get low frequency sub-images with quarter pixels, then attains the best matched sub-block by Full Search, at last the area of Reference Frame covered by the sub-block are searched to get the best matched point. Compared with traditional algorithm, experimental results show that this algorithm gives a significant improvement in accuracy for motion estimation and reduces the computational complexity, and nossesses strong robustness in different kinds of video sequences.

关 键 词：低频子图 运动估计 HAAR小波 块匹配算法 视频编码 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]