A research on registration and de-ghosting algorithm in image mosaic
Yang Yang, Qiong Xu, Zongliang Gan, Feng Liu · 2013
This paper presents methods for dealing with two problems which are common in image mosaic based on SIFT: time-consuming and ghosting. First, in order to solve the large amount of computation caused by SIFT keypoints, a fast registration algorithm based on effective registration area locating is proposed. The improved algorithm increases not only matching speed but also accuracy, since most of invalid keypoints have been rejected. Second, an improved frame difference algorithm based on HSI is used to detect ghosting area and a region cover method is used to eliminate ghosting. The experimental results show that the proposed automatic mosaic algorithm is fast, effective and has high value in practice.