Study On Image Stitching Based On Feature

Image stitching is an important research field of image processing. It is a technology that registers and fusions the series of images which are overlapped partly with each other, and finally generates a new integrate, seamless, wide-angle and high-resolution new image. At present, the image stitching technique is widely applied in various fields including micro-image analysis, deep space exploration, medical image analysis processing, virtual reality technology, remote sensing data processing, military reconnaissance, public security evidence collection and digital image ultra-resolution analysis etc.Image stitching technology is principally classified as the method based on the image features and the method based on the image region. Our researches belong to the former and the details are as follows.(1) A new detecting and extracting multi-scale feature points method based on empirical mode decomposition is proposed in this thesis. The first and the second intrinsic mode functions are obtained using the empirical mode decomposition algorithm. The Harris feature points are respectively extracted from the original image, the first intrinsic mode function and the superimposition of the first and the second intrinsic mode function by the Harris method. The candidate feature points are obtained with non-maxima suppression.(2) The feature points matching algorithm based on the normalized correlation coefficient method and the Golden Section method approach to above feature points is proposed in chapter 3. Firstly, we use the normalized correlation coefficient for feature points matching. And then, useful feature points are obtained by Golden Section method according to the correlation coefficient among the feature points. Experimental results show that the method can ensure the sufficiency and accuracy of matching points, and also independent on the selection of threshold in the matching process.(3) On the image fusion, a novel weighted technique based on cosine transform is proposed. Using the technique, not only the seamless stitching image is obtained, but also the phenomenon of ghosting in the overlapping stitching region is eliminated.(4) An efficient panoramic image stitching technique based on Euclidean distance between two matched feature points is proposed. The experimental results show that our method is effective and robust.