Study On Close-Range Stereo Image Matching Method Based On Nonsampled Contourlet Transform

Stereo image matching technique is a core technology in both realms of the Digital Photogrammetry and Computer Vision. One technology, which is used to get 3D data automatically by image matching technique, is paid attention to more and more. It is an important work to find out an efficient, stable and robust matching algorithm right now. For the close-range stereo image, the directions, shapes and proportional relations of objects in images are changed for the merging angles and the rotation angles caused by depth of field, shelter and images'error in position, so the accuracy of the matching about the close-range stereo image is low. Based on these, this paper will explore and study the methods about the close-range stereo image matching. The main results are as follow:(1) On the basis of analyzing the existing methods of the close-stereo images matching, a theory is advanced about the extracting method of the structural features of the close-range stereo image. The extracting method was found on Nonsubsampled Contourlet Transform. The new multiple dimensioned image geometry analysis method, which is called Nonsubsampled Contourlet Transform, would be used for stereo images matching. After the image is transformed through Nonsubsampled Contourlet Transform method, one low frequency information image and N high frequency information images would be gotten. In the high frequency ones, there are rich of detail information about the images, so we can extract the structural features of the image and construct the structural feature vector.(2) Through generalizing and analyzing the existing image matching similarity measure, propose an image matching similarity measure model that based on Nonsubsampled Contourlet Transform. By structural feature vector, construct the structural similarity measure. At the same time, taking into account the color image information, design the color similarity measure, and integrate the two composite similarity measures to improve the accuracy of the image matching.(3) On the support of the existing stereo image matching techniques, propose a stereo image matching method that based on Nonsubsampled Contourlet Transform. With a composite similarity measure model, and some constraints such as epipolar constraint, uniqueness constraint and continuity constraint, an example was given, which the data were got by Vehicle-borne spatial data collection system of the Key Laboratory of Virtual Geographic Environment, MOE, Nanjing Normal University. Compare several experimental results, seek the optimal scale and direction under Nonsubsampled Contourlet Transform, and ensure the precision and stability of the matching. This matching method is testified by those.(4) Finally, to compare the method in this paper with the gray correlation matching, the Hausdorff distance measure and SIFT operator matching, it is confirmed that close-range stereo image matching method based on Nonsubsampled Contourlet Transform proposed in this paper had high degree of accuracy and stability. In this situation, when contained certain angles about rotation, occlusion, or close range with a certain intersection angle stereo image pairs, the method can have higher matching accuracy.