Abstract: According to the problem of large amount of Scale Invariant Feature Transform(SIFT) algorithm, this paper constructs pyramid feature descriptor by using concentric squares, calculates every seed vector quickly by using recursion algorithm, and maintains the rotation invariant of algorithm by simply ranking the vector. It puts forward a matching search method which is suitable for the descriptor. Experimental results show that the algorithm can increase the matching speed, and can identify target effectively under rotate and zooming.

Key words: Scale Invariant Feature Transform(SIFT) algorithm, pyramid descriptor, seed vector, rotation invariant, half search method

摘要： 针对尺度不变特征变换(SIFT)算法计算量大的问题，提出一种基于快速SIFT特征提取的模板匹配算法。采用递推方法加速计算每个种子向量，利用向量排序来保持算法对旋转的不变性，并通过一种适用于该描述符的快速搜索匹配方法，提高算法的实时性。实验结果表明，该算法能提高匹配速度，并且能在旋转、缩放的情况下有效地识别目标。

关键词: 尺度不变特征变换算法, 金字塔描述符, 种子向量, 旋转不变性, 半数搜索法

CLC Number: 

• TP391.41