改进的立体像对稠密匹配算法  

Improved image dense stereo matching algorithm

作　　者：葛亮[1] 朱庆生[1] 傅思思 罗大江[1] 刘金凤[1] 

机构地区：[1]重庆大学计算机学院,重庆400030

出　　处：《吉林大学学报（工学版）》2010年第1期212-217,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：'863'国家高技术研究发展计划项目(2006AA10Z233);国家自然科学基金项目(60773082/F0205);重庆市自然科学基金项目(CSTC2006BB2229)

摘　　要：针对目前主流方法对图像中纹理单一区域匹配效果不佳的问题,提出了一种改进的立体像对稠密匹配算法。该算法首先利用区域增长技术找到图像中的纹理单一区域,然后将整个区域作为匹配基元以得到纹理单一区域的稠密视差图。相对于点基元,区域基元包含的信息更多,且在图像中不易重复出现,因此可以减少误匹配发生的几率。在国际标准测试图像上进行了实验,结果证明该算法的可行性与准确性。To obtain better disparity result of the less textured area of image, an improved dense stereo matching algorithm is proposed. First, the image is divided into several less textured or well textured areas based on region-growing strategy. Then, the dense disparity map of the less textured area is identified through the matching based on the whole region. Compared to pixel, a region contains more information and repeats seldom occur, thus reducing the probability of matching error. The proposed algorithm was tested with the international standard image data and compared with several existing algorithms. Experiment results show that the performance of the improved algorithm is better in handling less textured area and can be use in practice.

关 键 词：计算机应用 计算机视觉 立体匹配 基于区域的匹配 纹理单一 视差计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]