Abstract: SIFT features is invariance in zoom, rotate, scale, brightness variations. According to these advantages, this paper presents a novel dense disparity matching algorithm based on the reliable feature points of SIFT and edge region. The method determines SIFT features points in the edge and their disparity; The other points’ disparity are determined according to the principle of disparity gradient; A dense disparity map is generated. The experimental results show that the SIFT feature is to improve the accuracy of the disparity of feature points. And the disparity in some weak texture area has also been improved.

Key words: area-based stereo matching, Scale-Invariant Feature Transform（SIFT）, dense disparity map

摘要： 根据SIFT特征对旋转、尺度缩放、亮度变化保持不变性的特点，提出基于SIFT特征和边缘特征点的区域匹配方法。该方法确定符合SIFT特征的边缘为可靠特征点，并确定其视差；根据视差梯度原理确定其他点的视差，最后生成稠密的视差图。实验结果表明，SIFT特征的引入，提高了特征点视差的准确性，一些弱纹理区的匹配也有所改善。

关键词: 区域立体匹配, 尺度不变特征变换, 稠密视差图