Image Stitching Algorithm Research Based On SIFT

Image stitching technology is to match and amalgamate several images which maybe fromdifferent time, view or sensors and have overlapping parts to generate a panoramic image withgreater view. The panorama is seamless and ultra viewing angle or even360degree. Thetechnology which is good at solving the problem between large field of view and resolution is ahot research issue in the field of digital image processing, computer vision and so on. With thecontinuous development of computer technology, image stitching technology is widely used inspace detecting, heritage conservation, medical image processing, military surveillance, digitalvideo compression and virtual reality technology and other fields.Image registration is the core technology of image stitching. Image registration based onSIFT (Scale Invariant Feature Transform) is a typical image registration method based on pointfeatures. This topic proposes an improved image stitching algorithm through further study of theSIFT algorithm.The improved algorithm firstly establishes D2OG pyramid where the algorithm detects thezero-cross extremum points and generates SIFT feature points ultimately. Secondly, thealgorithm determines the matching points by using BBF (Best Bin First) algorithm based onkd-tree and then uses improved RANSAC (Random Sample Consensus) algorithm twice toeliminate mismatch and calculate the transform matrix. During the integration phase thealgorithm gets a smooth and seamless mosaic image by using weighted average method.Finally, the topic realizes the improved algorithm by programming. The experiments showthat: the improved algorithm ensures high precision and improves the speed of the algorithm atthe same time. At the end we analyse the deficiencies of the improved algorithm and proposefurther research directions.