点击次数：

论文类型：期刊论文

发表时间：2012-05-01

发表刊物：Journal of Information and Computational Science

收录刊物：EI、Scopus

卷号：9

期号：5

页面范围：1135-1143

ISSN号：15487741

摘要：Image alignment is a very important task in image processing. In this paper, we focus on the dense correspondence to deal with drastic changes and large displacement existing in the images. We introduce the more robust feature descriptor as the representation of the raw image pixel combining the optical flow model to build the correspondence. According to the statistical properties of dense descriptors, we propose robust function to reject outliers. In order to deal with occlusions, we propose a new method based on the robust regularization term which combines the varying support weight. The effectiveness of our method is borne out by abundant experiments. ? 2012 by Binary Information Press.