Abstract: Considering that the Scale Invariant Feature Transform (SIFT) algorithm extracts a great number of feature points, consumes a lot of matching time but with low matching accuracy, a fast image registration algorithm based on local significant edge features was proposed. Then SIFT algorithm was used to extract feature points, while wavelet edge detection was also used to extract image edge to establish feature points around the edge of the neighborhood characteristics, which filtered out points with a significant edge feature characteristic as significant feature points. A feature vector was formed by the shape-context operator and edge features. Euclidean distance was used as the match metric function to preliminarily match the feature points extracted from different images. Afterwards, RANdom SAmple Consensus (RANSAC) algorithm was applied to eliminate the mismatching points. The experimental results show that the algorithm effectively controlled the number of feature points, improved qulity of the feature points, reduced the feature search space and enhanced the efficiency of the feature matching.

Key words: Scale Invariant Feature Transform (SIFT), significant edge feature, wavelet edge detection, metric function, RANdom SAmple Consensus (RANSAC)

摘要： 针对尺度不变的特征变换(SIFT)算法提取的特征点数目多、匹配耗时长、匹配精度不高等问题,提出了一种基于局部显著边缘特征的快速图像配准算法。该算法利用SIFT算法提取待选特征点,同时用小波边缘检测提取图像边缘,建立特征点周围邻域的边缘特征,筛选出具有显著边缘特征的特征点,结合Shape-context算子和边缘特征形成特征描述向量,采用欧氏距离作为匹配度量函数对筛选出的特征点进行初步匹配,然后用随机一致性检验(RANSAC)算法消除误匹配点对。实验结果表明,该算法有效控制了特征点的数量,提高了特征点的质量,缩小了特征搜索空间,提高了特征匹配的效率。

关键词: 尺度不变的特征变换, 显著边缘特征, 小波边缘检测, 度量函数, 随机一致性检验