基于改进SIFT描述子的快速匹配算法  

A Fast Matching Algorithm Based on Improved SIFT Feature Descriptor

作　　者：钱冬云[1] 

机构地区：[1]浙江工贸职业技术学院信息传媒学院,浙江温州325003

出　　处：《浙江工贸职业技术学院学报》2015年第3期51-56,共6页Journal of Zhejiang Industry & Trade Vocational College

基　　金：2014年度浙江省高等学校访问学者教师专业发展项目(FX2014177);2014年度全国教育信息技术研究规划项目(146231964);2014年度温州市科技计划项目(R20140090)

摘　　要：针对尺度不变特征变换(SIFT)算法中存在描述子维度高、特征点信息冗余和配准运算量大等问题,提出一种改进的ECF-SIFT(eight circles features)算法。该算法通过高斯差分金字塔,确定特征点,并以环绕特征点的8个圆环为邻域构造64维的特征描述子,采用最近邻与次近邻之比对描述子进行匹配,最后用RANSAC算法对匹配点进行校正和消除误匹配。实验结果表明,在尺度缩放、尺度和旋转组合、视角变化、模糊变化和光照变化等条件下,ECF-SIFT算法的性能保持不变,并压缩了匹配时间,提高了匹配的准确率,算法的整体性能优于SIFT和SURF算法。An eight-circle-feature based scale invariant feature transform （ECF-SIFT） algorithm was developed to hurdle theproblems of the redundant feature point, large computation, and the high dimensionality of the scale invariant feature transform （SIFT）algorithm. The feature points were firstly determined by the difference of Gaussian. By utilizing eight concentric circles around the featurepoints, the 64-dimensional feature descriptors were created. The ratio of the first and the second closest distance was used tomatch the 64-dimesional vectors. Finally, the matching points were corrected by the RANSAC method to further remove the falsematching. The ECF-SIFT algorithm demonstrated the invariance of matching performance for rotation, scale scaling, the changes of illuminationchanges and smaller viewpoint, as well as the blur, which is superior to the SIFT algorithm and SURF algorithm with thehigher efficiency and better accuracy in image matching.

关 键 词：尺度不变特征变换 特征提取 特征描述子 RANSAC算法 图像配准 算法效率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]