Abstract: Stereo matching is one of the most active research subjects in computer vision. Aiming at the problems that traditional SIFT descriptor is subject to mismatch easily when the image has similar areas, and 200-dimensional description reduces the efficiency of Daisy descriptor algorithm, this paper presents a stereo matching algorithm based on the combination of SIFT and Daisy descriptor. The method uses SIFT algorithm to generate feature points, and then describes these feature points by Daisy descriptor for its good rotation invariant, and the disparity map is obtained according to the nearest neighbor matching based on Euclidean distance and seed region growing. Experimental results show this algorithm wins high matching accuracy and better performance when there are different levels of image geometric distortion or radiation distortion.

Key words: stereo matching, Scale Invariant Feature Transform（SIFT）, Daisy, feature descriptor

摘要： 立体匹配是计算机视觉领域最活跃的研究课题之一，针对传统SIFT描述符在图像存在多个相似区域时易造成误匹配和Daisy的匹配效率会因200维的描述符而降低的问题，提出一种SIFT和Daisy相结合的立体匹配算法。该方法利用SIFT算法生成关键特征点，利用Daisy描述符自身具有的良好的旋转不变性，对特征点进行描述，利用特征描述符欧氏距离的最近邻匹配和种子区域增长得到视差图。实验结果表明，该方法匹配精度高，速度快，在部分遮挡、视点变化引起的图像变形等问题上有更好的表现。

关键词: 立体匹配, 尺度不变特征变换（SIFT）, Daisy, 特征描述符