The Research On Stereo Matching Algorithm Based On Region-growth

The purpose of computer vision research is to acquire information by'visual', that is to say it obtains the three-dimensional information by processing the two-dimensional images, and then completes the recognition and understand for the shape movement and location of objects in the environment. With the development of the computer, this technology has been successfully applied to various fields.Stereo matching is a significant technology of the computer vision, and it has become a focal research project in various universities and enterprises in recent years. Stereo matching is the matching relationship between all the projected images which obtained form different perspective under the same features. Stereo matching technique is now widely used in the virtual scene reconstruction, medical treatment, navigation systems and other fields.This paper has done a lot of research work for the whole process of stereo matching technique. First, it proposes an improved clustering tree indexing algorithm which based on the vector quantization of characteristics, and this method is an effective search algorithm for the match points, which does not need to traverse the whole image and is different from the previous search strategy. Second, it introduces the concept of match strength for optimizing the seed points and improving the accuracy of matching, and then generates a dense disparity figure. This paper summarizes the basic knowledge of the feature extraction and matching, and then introduces the sift feature point extraction and the Harris corner in the scale space detection algorithm in detail, analyzes these experimental results from these methods finally. The paper primary studies the hash clustering tree indexing algorithm and using the method finds the matching points; in term of the stereo matching, introduces the principle of quasi-dense matching, and optimizes the seed points based on the traditional algorithm. The experiments show that the validity and accuracy are greatly improved by the improved stereo matching algorithm this paper proposed.