Affine Invariant Of Image Feature Matching Algorithm Research

Image feature extraction and matching is one of hot issues in the field of computer vision. Image features is affine invariant means that feature extraction keeps unchangeable as perspectives and camera parameters change. What’s more, it has superiority in algorithm robustness and application.Images in different views approximate affine transformation under certain conditions, it combines translation, rotation and scale transformation. This paper mainly studies with image feature of affine invariant and matching method. The image features are image skeleton and feature point. Skeleton, a good descriptor contains not only the geometric features but also topological information of image. It is a typical shape abstraction method. The research of affine invariant skeleton of image has great value in practical application. However, feature points are a tool to describe image and a key to recognize, because it contains rich information of position, gradient and class. In this paper, in the view of the above two features we give affine skeleton extraction algorithm and affine invariant feature points matching method in the sense of affine transformation. The specific research results include the following:(1) First, we achieve the skeleton matching by OSB algorithm in European space after, summarising and analysing the classic skeletonization method. Secondly, with further studying geometric properties and skeletonization algorithm of affine curve of affine space, we extend the affine invariant skeleton extraction method to geometry.(2) This paper proposes the improved SURF based on learning classical matching algorithm. In the case of guarantee the algorithm complexity, we use bidirectional SURF matching and affine invariance to extract affine invariant interest points to increase the correct matching number and rate under different perspectives, noise and illumination. So that the improved SURF algorithm has robustness and more practical value.