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

针对多视图匹配利用模板匹配精度低、计算量大等问题尤其是少纹理物体特征匹配问题,利用灰度或纹理特征进行图像匹配易产生误匹配等问题。提出基于边缘形状描述子,以实现多视点图像点匹配。首先,对边缘图像利用曲线凸性,将边缘分割为近似直线段组,将连接多个直线段的点称为关键点。然后,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:该方法有效地将匹配概率提高了23.64%,较SIFT匹配概率有较大提高;通过形状匹配实验证明提出算法在形状匹配中相对形状上下文匹配方法提高了6.67%。该方法对比例、平移具有不变性且对噪声等问题不敏感。

In order to solve the problem of low matching accuracy in the field of feature matching and point correspondence,a novel shape based on descriptor for point correspondence was proposed.First,based on the property of the curve convexity and the collinearity of the edge points,the edge map was divided into a group of roughly straight lines.The connection point of more than one straight line was defined as the key point.Then,all the straight lines around the key point formed a local shape patch.At last,by using the distribution of the edge points of the local shape patch related to the center of the shape patch,a novel descriptor was formed.Experimental results indicate that the proposed method can improve the correct matching probability about 23.64% than SIFT descriptor.According to the shape matching experiment,the results show that the proposed method can improve the correct matching probability about 6.67% than shape context method.The proposed descriptor is invariant to scale and translation,and it is not sensitive to the noise.