图像分割中运动估计研究  

Research on Motion Estimation Algorithm in Image Segmentation

作　　者：顾鹏[1] 袁杰[1] 卞春华[1] 

机构地区：[1]南京大学电子科学与工程学院,江苏南京210093

出　　处：《现代电子技术》2010年第20期60-62,66,共4页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(60501003)

摘　　要：为了改善运动估计中传统光流法处理的图像存在平滑性不好,边界信息丢失等缺点,提出了一种新的图像运动估计方法。该方法加入图像结构不变假设,降低光照变化的影响,并利用K-Mean聚类法对图像分割矫正,使得运动估计结果在保持平滑的基础上,更好地保留图像的边界信息。实验结果表明,该算法比传统光流法在运动边界问题的处理上更有好的改善。A novel approach is proposed for improving the loss of the smoothness and boundary information of the optical flow estimation for images in traditional motion estimation algorithms. Due to these problems, the assumption of the fixed image structure is added and the influence of illumination changes is reduced. Furthermore, the K-Mean clustering algorithm is used to restore more boundary information from the input images without losing extra smoothness in the stored images. The experiments show that the proposed method has a better result on the motion boundary estimation issues.

关 键 词：运动估计 光流法 K—Mean聚类 边界信息 

分 类 号：TN911-34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]