利用曲率和相关矩阵的角点检测算法

Corner detection with curvature and relation matrix

为提高定位准确性并抑制噪声,利用多方向Gabor滤波器,提出基于曲率和相关矩阵的角点检测方法.首先利用Canny边缘轮廓检测器提取图像的边缘并填充缺口;其次计算边缘像素点的曲率;然后利用多方向Gabor滤波器的虚部对原始图像进行平滑,对每个边缘像素及其邻域构造相关矩阵,利用相关矩阵的归一化特征值计算角点准则函数,将角点准则函数值与边缘像素点的曲率的乘积做为角点测度;最后利用非极大值抑制对候选角点进行筛选.分别在无噪声和噪声情况下进行实验,结果表明,与Harris,CPDA,He&Yung角点检测算法相比,该方法可以有效地抑制噪声,平均配准角点数提高了19.6%和25.6%,平均定位误差约降低了6.5%和9.2%.

In order to improve the positioning accuracy and suppress noise,the corner detection method based on curvature and relation matrix is proposed by using multi-direction Gabor filters.Firstly,Canny edge contour detector is used to extract edge map and fill gaps.Secondly,the curvature of edge pixels is calculate.Then the imaginary parts of Gabor filter is used to smooth the input image and build the relation matrix of edge pixel and its neighbour,the multiplication of the corner criterion function value computed by normalized eigenvalue of relation matrix and the edge pixel′s curvature is served as corner measure.Finally,use the non-maximum suppression to select the candidate corner.The experiment results contained with noiseand noise-free show that the proposed method can suppress noise effectively compared with Harris,CPDA,He&Yung algorithms,and the average matched corner number increases by about 19.6% and 25.6% respectively,and the positioning error reduces by about 6.5% and9.2%,respectively.