Line Matching Algorithm On Multispectral Images

In recent years, with the rapid development of computer vision technology, line matching technology is getting more and more attention.The inaccurate detection of. ending points brings a great challenge for matching lines, since corresponding lines may not be integrally extracted,thus affecting the establishment of the line descriptor. This paper presents a new line descriptor algorithm, combined with a post processing mechanism to eliminate mismatches based on improved RANSAC algorithm to improve the matching performance.In the process of line matching, to unify the line support area size, this dissertation will adopt the feature of line intersection as the line anchor support in order to establish a uniform size. In order to constrain the candidate line matching region, we first match the line intersection to find the suitable candidate matching line, which can improve the efficiency of the algorithm. In order to establish a stable line descriptor, the descriptor is combined with the sampling points in and around the line, and use the edge histogram (EOH) describe all points. However, there is no information about the main direction in the EOH descriptor method, and the accuracy of the registration of the gradient inversion phenomenon in the multispectral images is low. In this paper, the line direction is used as the main direction of the descriptor, so the line descriptor is robust, and the registration of the multispectral image is accuracy.In this paper, a post processing mechanism is proposed to eliminate mismatches. In order to improve the registration accuracy, the RANSAC mechanism, which is used to eliminate the mismatched keypoints, is applied to the line match through improved , which makes the algorithm get better results.In the last part, several algorithms are used to do registration experiment with 300 groups of images. Experimental results show that the proposed method has a significant improvement in the registration performance compared with the previous classical algorithms.