一种基于形状信息描述的图像特征匹配算法

Local shape patch based feature matching method

针对已有匹配方法匹配特征点少、图像匹配精度低的问题,提出一种基于边缘形状描述子的图像特征匹配算法。该算法首先利用曲线凸性将图像边缘分割为近似直线段组,再将连接多个直线段的点定义为关键点,将关键点周围直线段组定义为形状特征包,最后利用局部形状特征包中所有点集相对其几何中心的位置关系分布形成描述子。实验结果表明:该算法的图像匹配正确率均高于SIFT算法和Shape Context算法,且对噪声问题不敏感,可应用于基于Hough变换的物体检测。

Aimed at improving lower image matching accuracy due to fewer matching feature points in matching method in use,this paper proposes a novel image feature matching algorithm based on edge of the shape descriptor.The algorithm starts with dividing the edge map into a group of roughly straight lines,based on the property of the curve convexity of the edge points,proceeds to define the connection point of more than one straight line as the key point and form a local shape patch by using all the straight lines around the key point,and ends by forming a novel descriptor by using the distribution of the edge points of the local shape patch related to the center of the shape patch.Experiments of image matching and shape matching carried out by using the algorithm introduced in this paper presents indicate that the proposed method features a greater correct matching probability than SIFT algorithm and Shape Context algorithm and offers lower sensitivity to the noise,promising the application in object detection under the Hough transform.