基于局部保持匹配的改进图像匹配算法

Improved image matching algorithm based on locality preserving matching

针对传统图像匹配算法存在误匹配率高、匹配耗时长等问题,提出一种基于局部保持匹配(LPM)的改进图像匹配算法。该算法首先对图像进行双边滤波,减少噪声干扰,之后在尺度不变特征变换(SIFT)算法下提取鲁棒特征点,并利用欧氏距离进行粗匹配得到初始匹配集;然后利用LPM算法对初始匹配集中具有相似局部邻域结构的特征点进行保留,剔除不符合条件的匹配对,得到二次匹配集;最后利用改进的随机抽样一致性(RANSAC)算法建立局部最优模型,并对二次匹配集进行误匹配点剔除,得到精匹配集,实现高精度匹配。通过采用Oxford标准数据集与真实图像进行仿真的实验结果表明,与其他几种算法相比,改进的图像匹配算法在图像发生旋转、缩放、尺度以及光照等变化时匹配精度与匹配实时性均有显著提升,因此,改进算法具有良好的匹配性能。

Aiming at the problems of high false matching rate and long matching time in traditional image matching algorithms,this paper proposed an improved image matching algorithm based on Locality Preserving Matching(LPM).Firstly the algorithm performs bilateral filtering on image to reduce noise interference,and extracts robust feature points under the Scale Invariant Feature Transform(SIFT)algorithm.Then,it uses the Euclidean distance for rough matching to obtain initial matching set;Secondly the LPM algorithm is used to retain feature points that have similar local neighborhood structure in initial matching set,and the unqualified matching pairs are eliminated to obtain secondary matching set;Finally,the improved Random Sample Consensus(RANSAC)algorithm establishes local optimal model,eliminates mismatched points,and obtains fine matching set that has a high matching precision.Experimental results using Oxford standard data set and real images show that compared with other algorithms,the improved algorithm has significantly improved matching accuracy and real-time matching when the image is rotated,zoomed,scaled,and illuminated,which has a good performance.