基于复杂运动模型的运动估值算法研究  

Study on Motion Estimation Using Complex Motion Models

作　　者：薛向阳[1] 周强[1] 吴立德[1] 

机构地区：[1]复旦大学计算机科学系

出　　处：《通信学报》1998年第3期62-67,共6页Journal on Communications

基　　金：自然科学基金;国家863项目;中国博士后基金

摘　　要：传统的块匹配运动估值算法（ＢＭＡ）不适合对下列情形进行运动估计：旋转、伸缩（ｓｃａｌｅ）、形变（ｄｅｆｏｒｍａｔｉｏｎ）、卷曲（ｗａｒｐｉｎｇ）、错切（ｓｈｅａｒ）和不规则的伸展（ｕｎｅｖｅｎｓｔｒｅｔｃｈｉｎｇ）等，因而在实际应用中常常出现方块失真、沿运动边缘方向预测性能差等现象。为克服上述缺陷，本文研究了基于多种运动模型的运动估计算法，提出了一种自适应选择运动模型的算法。为了估计模型参数，本文实现了基于模拟退火的运动参数估计算法。实验表明本文算法与传统ＢＭＡ算法相比较，较好地提高了帧间预测增益。It is well known that traditional block matching motion estimation algorithm (BMA) is not suitable to estimate the following motion cases:rotation,scale,deformation,warping,shear,uneven stretching,etc.,so blocking effects and poor predicting performance along edge direction always appear in image sequence coding.In order to get rid of the above shortcomings,we study on some complex motion models,such as affine transform and the second order geometrical transform,propose an adaptive model selection algorithm,and introduce simulated annealing method to estimate the parameters of the complex models.Results show the annealing algorithm can estimate motion parameters successfully,and the interframe prediction performance is improved greatly.

关 键 词：运动估值 视频编码 模拟退火 图像处理 

分 类 号：TN919.8[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]