Abstract: Based on the cellular neutral network(CNN) mathematical model, a new
method of color image edge detection is proposed. It inherits the advantages of CNN, and fills
up a gap that the existing algorithms cannot directly detect the color image edge. This method
takes advantage of the color information in the image to measure the differences between
pixel by the Euclidean distance, allowing the CNN equation for operation in RGB color space.
To make a theoretical analysis and robustness research for CNN template, a CNN robustness
theorem is proposed that achieves the functional requirements of the color image edge
detection, and provides the analytical criteria for designing the corresponding CNN templates
parameters. The experimental results show that this algorithm is effective for the color image
edge extraction. A quantitative evaluation validates the advantages of the accurate edge
detection positioning.

Key words: edge detection, cellular neural networks (CNN), robustness, template design