Research On The Application Of Region Detection Operator And Image Registration

Image registration is a very important technology in modern information processing, and it has been widely applied in computer vision, artificial intelligence and other fields. First, this paper introduces the basic concept of image registration, the purpose and significance of image registration and its research status at home and abroad, and we has researched the theoretical basis of image registration, and has introduced the mathematical definition of image registration, image transform model and common registration method, and will take the image registration method based on features as the research focus of this article.In the study of the local characteristics of the image we focus on the region detection operator. Region detection operator is stable region in the image which has significant features with the good affine invariance. The comprehensive properties of MSER operator is the best. And classical SIFT algorithm and its accelerating version of SURF algorithm both have the defect of poor affine invariance, This paper is aiming at the above problem, in order to solve the disadvantage of SIFT and SURF algorithm as the main object of study, in depth analysis of the algorithm MSER, SIFT and SURF, presented a new method of MSER combined with SIFT and SURF respectively to realize the complementary advantages. The main research contents of this paper include the following aspects:1. MSER, SIFT and SURF algorithm has been studied and implemented successively, and deeply analysis their respective advantages and disadvantages:The matching effect of SIFT algorithm is the best but are poor in computing speed and affine invariance, The SURF algorithm is faster but the matching effect is not as good as SIFT algorithm, and MSER have good affine invariance.2. We proposed to MSER combined with SIFT and SURF respectively to realize complementary advantages. Due to SIFT and SURF algorithm are a complete set of image registration algorithm, which includes feature point extraction, description and feature points matching, While MSER is only to extract certain regions in images, therefore the key of the new method is how to combine the two algorithms together, After in-depth study of each algorithm characteristics, we finally devised a better combination method.3. We performed experiments on the new image registration method and the original one in the unified test set, after comprehensive comparison and analysis on the experimental data, we learned that the processing time of the method of MSER combined with SIFT algorithm is too long and its utility is poor, While MSER combined with SURF algorithm realize the complementary advantages, which combines the SURF algorithm fast processing speed and the advantages of the affine invariance of MSER.4. Through the analysis of the characteristics of infrared image, we learned that the MSER-based detection algorithm is suitable for infrared image processing, and then put forward the mosaic of the infrared image using MSER combined with SURF algorithm.