Research On Image Matching Algorithm Based On Harris Scale Invariant Feature

Image matching is a fundamental problem in remote sensing, medical, computer vision and many other fields. In the field of remote sensing, image matching is essential as a key step in realizing image fusion, changing detection, image correction, image mosaic, and other applications. Because of the remote sensing information and a wide range of applications, automatic matching has been a goal pursued by the people.This paper adopts an image matching algorithm based on features. Our work in this dissertation is as follows: First, the paper does some contrastive experiments on three common point feature detection methods. According to the repeatability index, the experimental results show that Harris scale invariant detector with valid parameter obtains the best performance. Second, this paper mainly describes SIFT matching algorithm based on features, and introduces SIFT feature description which is the best feature descriptor to Harris scale invariant feature description. The results demonstrate this method is effective on rotation invariant, illumination invariant and scale invariant in matching images, and has a certain stability to the noise. This method improves the stability of the matching.In the matching algorithm, the Euclidean distance between feature vectors is considered to be measured the similarity between features. In order to improve the accuracy of image matching and remove most of the false matches, this paper uses a bilateral matching strategy based on the nearest-neighbor algorithm. First, Harris feature vectors in two images are extracted; Second, after taking each feature point of the first image, we use exhaustive search to find the two points of the second image which are second closest in Euclidean distance, if the nearest distance divided by the second one is less than the threshold, then this point of the first image is viewed to be matched the point of the second image which has the nearest distance. Finally, the matched points of the second image in the previous step are calculated the corresponding points of the first image. The paper does some experiments on image matching methods based on SIFT features and Harris scale invariant features respectively. According to matching accuracy rate, the experimental results show that the image matching algorithm based on Harris scale invariant detector obtains higher matching accuracy rate and more stable performance,therefore,it is suitable for the case which requires higher matching accuracy rate.