Research On Image Matching Algorithm For Underwater Environment

With the development of the world economy and the increasing of population,land resources have been exploited and there is little left,therefore, people’s attention has to be turned to the marine resources, exploration and target location of deep sea and target tracking technology are getting more and more attention all over the world, human is unable to get to the dangerous and complex deep sea area, so underwater robot has become the best choice and technology in the field of robot vision has become the key of the research. The core of machine vision technology is to process the collected images, whereas image matching is the core technology of image processing, which is a process to find correspondence points between two or more images by appropriate algorithms. Because of the problem of high noise, weak light, low visibility and inconspicuous texture of object in the water environment, how to accomplish extracting and matching accurately and quickly for feature of underwater image is the focus of this paper.This paper starts from the image matching theory, and makes detailed introduction for Moravec、SUSAN、Harris three kinds of feature extraction algorithms, then it is pointed out that the feature points of Harris extraction algorithm has a better performance. Then the six steps of the feature points extraction of SIFT algorithm has been discussed and the performance of SIFT feature descriptor is analyzed. According to the problem of SIFT algorithm, such as feature points are too many, dimension of descriptor is too high, bad real-time performance and low matching precision, a new improved algorithm is proposed. The Harris operator is introduced into multi-scale space, and corner points will be detected as the feature points of the image by Harris in the DOG scale space, then using a circular neighborhood window and the character of gradient to reduce the dimension of feature descriptor, thereby the rate of feature extraction has been speeded up effectively. At the same time, the bidirectional matching is introduced in the matching process to improve the accuracy. Finally, the test experiment and the simulation of underwater images are carried out in the MATLAB R2009b software using the SIFT algorithm, SURF algorithm and the improved algorithm in this paper. Results of the experiments show that the improved algorithm has good real-time performance, high matching rate, well anti noise, and is more suitable for processing the image under water.