基于透视不变二值特征描述子的图像匹配算法

Perspective invariant binary feature descriptor based image matching algorithm

针对基于局部特征的图像匹配算法普遍存在对透视变换顽健性差的缺点,提出了一种新的二值特征描述子PIBC(perspective invariant binary code),提高了图像匹配算法的透视变换顽健性。首先,在提取金字塔图像FAST特征点的基础上,利用Harris角点响应值去除非极大值点和边缘响应点;其次,通过模拟相机不同视角成像之间的透视变换,对单个FAST特征点生成不同视角变换下图像的二值描述子,使描述子具备描述不同视角图像中同一特征点的能力。实验结果表明,算法在提高描述子透视不变性的同时时间复杂度与SURF算法近似。

Current local feature based image matching algorithms are usually less robust to image perspective transfor- mation. Aiming to solve this problem, a new perspective invariant binary code （PIBC） based image matching algorithm is proposed. Firstly, FAST comers are detected on the pyramid images, those comers with non-maximum Harris comer re- sponse value and the edge points are further eliminated. And then, by simulating the perspective transformations of im- ages taken from different viewpoints, a single FAST comer is described with binary descriptors trader different viewpoint transformations, which makes the descriptor could describe the identical feature point on different perspective transform images. Experimental results show its robustness to image perspective transformation, while its complexity is similar with SURF.