Aimed at the problems in traditional image matching algorithms having long image matching time and high mismatching rate,an image matching algorithm of improved FAST (features by accelerated segment test)and FREAK (fast retina keypoint)is proposed.Firstly,in this algorithm,the quantity of the pixel points within a circular neighborhood is continuously altered and compared with other FAST pixel templates,so that a method for extracting the FAST-9 feature points is established.Then its FREAK lo-cal invariant feature descriptor is calculated to generate feature vectors.Finally,RANSAC consensus sie-ving is performed to eliminate mismatching points.The experimental result shows that compared to SIFT and BRIEF algorithms,the proposed algorithm will be able to reduce image matching time and improve image matching accuracy to a certain extent and will have a better robustness of rotation deviation,scale deviation and lightening deviation.%针对传统图像匹配算法匹配时间较长、误匹配率较高的问题,提出一种改进的FAST和FREAK的图像匹配算法.该算法首先在圆形邻域上不断改变像素点个数,并与其他 FAST像素模板进行对比,从而建立 FAST-9 特征点提取方法;然后计算其 FREAK局部不变特征描述符,生成特征向量;最后通过 RANSAC 一致性筛选剔除误匹配点.实验结果表明,本文算法与 SIFT、BRIEF算法比较,图像匹配时间缩短且图像匹配精度有一定的提高,并且对图像的旋转差异、尺度差异和光照差异都具有较好的鲁棒性.
展开▼