Edge Detection Based On Cellular Neural Networks

Image edge is one important feature of the image, also is an important foundationof computer vision and pattern recognition. Recently, with the development ofcomputer vision and machine vision, edge detection which is an important problem inimage processing receives many researchers’ interest. Because Cellular NeuralNetworks has two-value output characteristic and the parallel processing capability, soit is ideal to use CNN detect image edge. This paper involves three main researchingaspects, which are image denoising before edge detection and gray image edgedetection and color image edge detection. The network’s requires have been reachedfrom the analysis of CNN state and output. Base on these requires, the gray imageedge detection CNN model and color image edge detection CNN model have beenproposed. The template parameters of CNN have been trained out by PSO. Using thedesigned CNN to detect gray image edge and color image edge, the performanceimproves greatly. The main contents are as follows:①The CNN models for noise removal of binary image and gray image haverespectively been established. Then the optimal template parameters of CNN aresearched out from the multiple-dimensional space by particle swarm optimization(PSO) algorithm. Simulation results show that optimized CNN has an advantage on theeffectiveness and speed for denoising. This indicates that CNN has a certain noiseimmunity when used to detect image edge.②The analysis and comparison of the classical edge detection algorithms are madein this paper. In order to overcome the problems in the existing edge detection, thespace-invariant template CNN and the space-varying template CNN are proposed inthis paper. Then the template parameters of CNN are trained out by PSO algorithm.Lastly, the performance of designed CNN has been evaluated. The simulation resultsshow that the edge detected by optimized CNN is more accurate and better inperformance.③In the color space, based on the human eye color perception limits, this paperestablishes a color image edge detection CNN model. Firstly, this paper brieflyintroduces the color space of the color image, and analynize the color gradient in thecolor space. Then the human vision perception limits model is proposed, and based onthis, the color image edge detection CNN model is established. The template parameters of designed CNN are optimized by TVSC-PSO algorithm. From the simulation results,the convergence rate of the optimized CNN is100%, and the edge image is betterdetected by the optimized CNN.