形状自适应各向异性微分滤波器边缘检测算法

Edge detection algorithm using shape-adaptive anisotropic differential filter

传统参数固定的微分滤波器难以精确检测图像中不同类型的边缘,并存在噪声敏感的不足,为此,提出了一种基于形状自适应各向异性微分滤波器的边缘检测算法。利用图像的微分自相关矩阵构建一种反映边缘像素类型的度量准则,并建立度量与各向异性高斯方向导数(anisotropic Gaussian directional derivative,ANDD)滤波器各向异性因子之间的映射,实现ANDD滤波器的形状自适应控制,从而能精确地提取不同类型边缘的强度映射。同时大尺度的ANDD滤波器增强了边缘强度映射的噪声鲁棒性。实验结果证明,在无噪声情况下,所提算法的边缘品质因子(pratt figure of merit,FOM)分别比Canny边缘检测算法、基于Gabor的边缘检测算法和基于测度融合的边缘检测算法高23.3%、14.5%和9.5%。在含噪声情况下,则分别高41.7%、29.7%和12.0%。

Abstract： The traditional differential filter with fixed parameter has defects that it is difficult to precisely detect the edges of different type and is noise-sensitive. Therefore, an edge detection algorithm based on the shape-adaptive anisotropie differential filter is proposed. Using the differential autocorrelation matrix constructs a measure which can reflect the type of edge in image. Then, a map function from the measure to the anisotropic factor of anisotropie Gauss{an directional derivative （ANDD） filter is designed to achieve the goal of adaptively controlling the shape of ANDD filter and progress to extract the precise edge strength map of different type. The ANDD filter with large scale improves the robustness of the edge strength map to noise. The experimental results show that the pratt figure of merit （FOM） of the proposed algorithm is respectively improved by 23.3%, 14.5% and 9.5% compared with the Canny edge detection algorithm, the Gabor-based edge detection and the measure fusion-based edge detection algorithm under noise-free situation, and is respectively improved by 41.7%, 29.7% and 12.0% under noisy situation.