文档介绍:河北大学硕士学位论文基于区域增长的立体匹配算法的研究姓名:薛丽娟申请学位级别:硕士专业:通信与信息系统指导教师:赵杰 2011-06 摘要 I 摘要计算机视觉研究目的是通过“视觉”获取信息,即利用计算机对采集到的二维图像进行处理,获得三维信息,从而完成对周围环境中物体的形状、运动及位置的识别和理解。随着计算机的发展,这一技术已成功应用到各个领域中。 立体匹配是计算机视觉的一项关键技术,也是近年来各个高校、企业的重点科研项目。立体匹配是指在不同的视角下获得的同一景物的各个投影图像间的匹配关系。立体匹配技术被广泛的应用在虚拟场景的重建、医学处理、导航系统等领域。本文对立体匹配技术的整个过程都做了大量的研究工作,首先提出了一种改进的基于特征向量量化的聚类树索引算法,此方法不同于以往的搜索策略,不需要遍历整幅图片,是一种有效的匹配点搜索算法。其次引入了匹配强度这一概念来优化种子点,提高匹配精度,生成了致密的视差图。本文总结了特征提取及匹配的基础知识,详细介绍了 SIFT 特征点提取及尺度空间的 Harris 角点检测算法,并对实验结果进行了分析。重点研究了哈希聚类索引算法寻找匹配点;在立体匹配方面介绍了准稠密匹配的原理并在传统的算法上优化种子点。实验证明,经过改进的立体匹配算法在有效性和准确性上都有很大的提高。 关键词 计算机视觉 立体匹配 特征点提取 哈希索引算法 Abstract II Abstract The purpose puter vision research is to acquire information by ‘v isual’, that is to say it obtains the three-dimensional informa tion by processing the two-dimensional images, and pletes the recognition and understa nd for the shape movement and location of objects in the environment. With the developm ent of puter, th is technology has been essfully applied to various fields. Stereo matching is a significant technology of puter vision, and it has e a focal research project in various universities an d enterprises in recent years. Stereo matching is the matching relationship between all the pr ojected images which obtained form different perspective under the same features. Stereo matc hing technique is now widely used in the virtual scene reconstruction, me dical 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 propose s an improved clustering tree in dexing algorithm which based on the vector quantization of charac teristics, 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 genera