Study On Motion Estimation In Digital Image Stabilization

Nowadays, there are very wide applications of digital image stabilization in national economy and national defense. One of the research key points is how to eliminate or reduce the unwanted motions between image frames in a fast and robust way. Considering the problems of translation and rotation motions in digital image stabilization, the thesis focuses on studying the motion estimation of such problems. The most contents and innovations are:First, the thesis enumerates estimation methods of translation motion, analyses the principle and the important affecting factors of block matching method, studies the improving ways of other algorithms, and proposes a new translation motion estimation algorithm. It combines bit-plane matching with diamond strategy, and improves the full search block matching method using full image frame in three approaches: data that used in matching,matching criterion and search strategy. Compared with the original method, our algorithm is much faster while it maintains accuracy.Second, the thesis presents estimation algorithms of complex motions including translation, rotation and scale, and analyses the difficulty caused by large rotation. With the idea dealing with rotation and other types of motion separately in image registration, a new image rotation angle estimation algorithm is proposed. The algorithm improves the voting procedure of traditional angle histogram method, and can estimate the rotation angle fast and robustly when the angle between two images is larger than five degrees. It can apply not only in digital image stabilization, but also in image registration and image mosaic.Experimented with many data sets, the results demonstrate the effectiveness of ours algorithms. Based on such innovations, three papers have been published and accepted, including two conference papers and one journal paper.