结合FAST-SURF和改进k-d树最近邻查找的图像配准  

Image matching algorithm combining FAST-SURF and improved k-d tree nearest neighbor search

作　　者：陈剑虹[1] 韩小珍[1] 

机构地区：[1]西安理工大学机械与精密仪器工程学院,陕西西安710048

出　　处：《西安理工大学学报》2016年第2期213-217,252,共6页Journal of Xi'an University of Technology

基　　金：陕西省自然科学基础研究计划资助项目(2012JM8006);陕西省教育厅科研计划资助项目(2013JK1049)

摘　　要：针对两图像之间存在平移和旋转变化的图像匹配,提出了一种结合FAST-SURF和改进k-d树最近邻查找的图像配准算法。该算法首先用FAST(加速分割检测特征)检测器进行特征点提取,然后根据特征点周围邻域的信息生成SURF(快速鲁棒特征)描述子,采用一种改进的k-d树最近邻查找算法BBF(最优节点优先)寻找特征点的最近邻点及次近邻点,接着进行双向匹配得到初匹配点对,最后利用RANSAC(随机抽样一致性)算法消除误匹配点,findHomography函数寻找单应性变化矩阵,从而计算出图像间的相对平移量和旋转量。实验结果表明,该算法平移参数的最大误差为0.022个像素,旋转参数的最大误差为0.045度,优于传统的SURF图像匹配算法,实现了图像的快速、高精度配准。An algorithm combining FAST-SURF and improved k d tree nearest neighbor search is proposed to solve the matching problem of translation and rotation changes between two images. Feature points are first extracted using FAST （Features from Accelerated Segment Test） corner detector, and then SURF （Speeded Up Robust Feature） descriptors are generated based on the feature points around the neighborhood information. And an improved k-d tree nearest neighbor search algorithm BBF （Best Bin First） is adopted to find out the feature highlights of two nearest neighbor points. The preliminary match point is obtained by bidirectional matching, and finally RANSAC （Random Sample Consensus） algorithm is adopted to eliminate false matching points, with findHomography function used to find transformation matrix to calculate the relative amount of translation and rotation of the two images. Experimental results are superior to traditional SURF image matching algorithm with the maximum error of the algorithmic translation parameter being 0. 022 pixels, and the maximum error of rotation parameters is 0. 045 degree, thus achieving a fast and accurate precision of the image registration.

关 键 词：图像匹配 FAST-SURF算法 BBF 双向匹配 RANSAC 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]