一种基于小波分解和分水岭变换的视频对象自动分割算法  

Automatic video object extraction algorithm based on wavelet and watershed

作　　者：杨高波[1] 张兆杨[2] 余圣发[3] 

机构地区：[1]湖南大学计算机与通信学院 [2]上海大学通信与信息工程学院 [3]国防科学技术大学电子科学与工程学院

出　　处：《通信学报》2005年第3期7-13,20,共8页Journal on Communications

基　　金：国家自然科学基金资助项目(60172020);湖南大学校基金资助项目

摘　　要：提出了一种基于时空分割融合的视频分割算法。时间分割采用变化检测,其关键的阈值选取通过直方图分析得到。空间分割时,先对图像进行二级小波分解,在低分辨率图像上进行分水岭变换,它可以有效克服噪声的影响并有效地减少过分割。实验结果表明,它可比 COST211AM 参考分割取得更好的分割结果。A spatio-temporal video object segmentation algorithm was proposed. Temporal segmentation was based on change detection. Its key was the selection of appropriate threshold, which was obtained by histogram analysis. For spatial segmentation, discrete wavelet transform was first conducted, and then watershed was performed on the low-resolution image. A region-based classification scheme was utilized to fuse the temporal and spatial segmentation. The proposed approach can get rid of the influence by noise and effectively overcome the “over-segmentation”. Experimental results demonstrate that it can achieve more desirable results than COST211 AM.

关 键 词：视频对象提取 小波分解 区域合并 分水岭变换 区域投影 

分 类 号：TN911.21[电子电信—通信与信息系统]