基于相对变分的边缘检测

Edge detection based on the relative variation regularization

边缘检测是机器视觉与图像理解中的基础问题,准确地提取轮廓有助于提高后续操作的质量和效率。在深入分析图像纹理结构、噪声及细胞神经网络算法(cellular neural network,CNN)的基础上提出一种基于相对变分正则化的细胞神经网络边缘检测方法。首先采用相对变分正则化的方法将图像的纹理进行平滑,去除纹理以及噪声对边缘提取的影响;然后再用标准的细胞神经网络算法对平滑后的图像进行边缘检测。实验结果表明:算法与Canny、CDCNN等算法相比,在没有重新设计新的复杂CNN模板参数的情况下,针对具有复杂纹理及含有一定量噪声的图片进行边缘检测,算法能得到更好的检测结果。

Edge detection is a basic problem of machine vision and image understanding,extracting contour accurately can improve the quality and efficiency of subsequent operations.This paper presents a method of Cellular Neural Network （CNN） based on relative variation regularization after analyzing image texture,noise,and cellular neural network algorithm.Firstly,in order to remove the impact of texture and noise,using relative variation regularization to smooth a image.Secondly,choosing the standard Contour Detect Cellular Neural Network（CD-CNN） to obtain the edge detection for the smoothed image.And at last,the method proposed in this paper yields significant performance improvements over the Canny,and over the CD-CNN without designing a new CNN template parameter for a certain texture and noise image.