Edge Detection for Color Image Based on CNN

Abstract

According to cell neural network (CNN) model, this paper presents a new color image edge detection algorithm which inherits the advantage of CNN and makes up the disadvantage of CNN method which can not directly deal with color images. This algorithm takes full advantage of color image information, through euclidean distance measures the difference between pixels, so that CNN equation can be carried out to compute in the RGB color space. According to theoretical analysis and robustness studies for CNN template, this paper also presents a CNN robustness theorem which can implement edge detection for color image. This theorem provided analytical criterion for designing parameters of the corresponding CNN templates. The experimental results show that the proposed approach can effectively carry out edge extraction and quantitative assessments verify the advantage of CNN edge detection which can locates accurately.