Research On Fast Gray Image Matching Algorithm

Image matching is an important research direction in computer vision andimage processing field. Now the complexity of the image brings difficulties to thematching process, so selecting a fast and suitable matching algorithm increasinglybecome the focus of the current image matching, and has both important theoreticalsignificance and application value. Image matching methods mainly contain twocategories: one is based on gray, the other is based on features.In this thesis, there are mainly two types of image matching methods beingresearched: method based on the gray-value of pixels and method based on thefeatures. After deeply studying on the current matching algorithms and the analysisof related works, the new fast image methods are proposed, which can respectivelyimprove the shortcomings of the two types of matching algorithms. The maincontent of this thesis is shown as follows:1. In terms of matching algorithm based on image gray value, a new algorithmof fast template matching based on normalized cross correlation(NCC) is proposed.In this paper, area centroid and histogram information are used to select thematching candidates, then using NCC to get the final result. Additionally, thesearching starts from the image centroid, instead of usual upper left corner points.The new algorithm not only reduces searching and matching time, but alsoguarantees the matching accuracy according to the distribution of key features.Therefore, the new algorithm improves the speed under the premise of ensuring thematching effect.2. In terms of matching algorithm based on image character, a new algorithmof fast image matching based on Scale Invariant Feature Transform (SIFT) isproposed. Firstly, a new approach to generate the SIFT feature descriptor is used toreduce the dimension of feature vector, which makes the dimension from128downto64by using circle uniqueness. Meanwhile, the new formula is used to describethe distance between two features. The improvement measures mentioned above areaimed at reducing matching time. Secondly, bilateral matching strategy based onthe nearest neighbor algorithm is also used to improve the matching accuracy,combined with Random Sample Consensus (RANSAC) algorithm to eliminatemismatch point.