Abstract: Conventional color edge detection algorithms may detect noise as edges when improving the accuracy of edge detection.While improving the robustness of noise, some edges will be suppressed as noise, resulting in the loss of some edge information.To resolve the conflict between edge detection accuracy and noise robustness in conventional color edge detection algorithms, a new color edge detection algorithm using adaptive Anisotropic Gaussian Directional Derivative(ANDD) is proposed.Through differential autocorrelation matrix of the color image, a measurement criterion reflecting the type of edge is constructed to adaptively determine the shape of the ANDD filter at each pixel, so that the edge features of the different types can be accurately extracted.The ANDD filter bank is used to smooth the image to extract the ANDD features on three channels.Based on this, the optimal fusion weight is determined by Singular Value Decomposition (SVD), and the ANDD features of the three channels are fused to improve the intensity of the color edges.The experimental results show that the Pratt Figure Of Merit (FOM) of the proposed algorithm are 0.849 6 and 0.791 4 in noise-free and noisy environments, respectively.Compared to the color Canny, RCMG-MM, and FRPOS algorithms, the proposed algorithm has better noise robustness while maintaining high edge detection accuracy.

Key words: color edge detection, anisotropic Gaussian, autocorrelation matrix, Singular Value Decomposition(SVD), feature fusion, directional derivative filter

摘要： 传统彩色边缘检测算法在提高边缘检测准确性时可能将噪声检测为边缘，而在提高噪声鲁棒性时会将部分边缘当作噪声进行抑制，导致部分边缘信息丢失。为解决传统彩色边缘检测算法在边缘检测准确性与噪声鲁棒性之间的矛盾问题，提出一种基于自适应各向异性高斯方向导数（ANDD）的彩色边缘检测算法。通过彩色图像的微分自相关矩阵构建反映边缘类型的度量准则，以自适应地确定每个像素处ANDD滤波器的形状，从而准确提取不同类型的边缘特征，采用ANDD滤波器组对图像进行平滑处理，提取在三个通道上的ANDD特征。在此基础上，利用奇异值分解得到最优融合权值，并融合三个通道的ANDD特征，以增强彩色边缘强度。实验结果表明，该算法在无噪声和含噪声环境下的Pratt品质因子分别为0.849 6和0.791 4，与彩色Canny、RCMG-MM和FRPOS算法相比，在保持较高边缘检测准确率的同时具有较优的噪声鲁棒性。

关键词: 彩色边缘检测, 各向异性高斯, 自相关矩阵, 奇异值分解, 特征融合, 方向导数滤波器

CLC Number: 

• TN911.73