Study On Feature Matching Based Image Stitching Technologe

With the rapid development of computer technology and digital image processing technology, image stitching has become an important aspect of image rendering technology and also is a popular subject about image processing technology research. Because of constraints of image-capturing hardware, the image obtained usually can't satisfy the requirements of dimension and precision; therefore, computer-aided image processing algorithm becomes necessary to expand dimension and improve precision. This paper attempts to do a comprehensive research on image stitching technology and meanwhile focuses on feature-based image stitching technology. Based on fundamental theoretical analysis of computer vision and pattern recognition technology in combination of the key technology of feature extraction, matching, integration and so on, a panoramic image stitching system is designed and developed.The key aspects and efforts through this paper include:(1) A detailed introduction to panoramic stitching technology background knowledge, prevailing research circumstance, subject, and various methodologies in view of the importance of this technology in real application.(2) A thorough introduction of design flow of the panoramic image stitching system. Based on the basic principle and methodology of feature extraction, matching and integration on image stitching, an appropriate algorithm is conceptualized on the need of this research and according technical roadmap is designed and implemented for image stitching.(3) This paper proposes to use SIFT(Scale Invariant Feature Transform)algorithm for image extraction and matching because of its relatively high detection capability commonly for circumstances that spherical video surveillance equipments usually have distinct focus, complex image capturing environments, frequent change of rotation, prospect, brightness, and obstacle. Experimental results have proven that this method's high precision and applicability for stitching of image captured with spherical video surveillance equipments.(4) In this paper, a coarse-to-fine matching strategy is adopted: Euclidean distance-based methods is used for preliminary characteristics match, and then RANSAC(Random Sample Consensus)algorithm and the least square method in the inner point scope are utilized to accurately estimate the transformation matrix between images. Experimental results show that the RANSAC algorithm for purification can help the system to reduce the unnecessary computation of mismatch points significantly and filter out the pseudo-match point effectively, thus enhance the computation efficiency and correctness.(5) Image fusion techniques is adopted in this paper to post-process the panorama after stitching in order to eliminate stitching vestige. Taking into consideration the balance between lowering of algorithm complexity and improving of fusion effect, evolutional fusion algorithm is taken advantaged of to process image overlap region for smooth transition to achieve seamless image stitching effect.(6) Feature based image sequence stitching software module is developed with Microsoft VC++. Processing and computation over the image helps not only capture and obtain the feature parameters and the matching circumstance of the current image in a fast pace, but also implement well the Panoramic stitching of image sequence with the same overlap region.