Color Image Segmentation Of Multi-channel Pulse Coupled Neural Networks

Pulse coupled neural network( PCNN) is suitable for gray image processing, color image can’t be directly processed. Only an external input is received by traditional PCNN, because it is a single-channel model. In this paper, on the basis of previous studied PCNN model, multi-channel pulse coupled neural networks of simultaneously receiving multiple images is improved by extending PCNN( M-PCNN). Maximum entropy criterion is combined to achieve color image segmentation. The main contents of the article in the following areas:First, studied the works principle、basic applications of PCNN and the evaluation criteria of image segmentation results. Through comparative analysis of the experimental results, maximum entropy criteria is selected as the objective evaluation criteria of image segmentation results.Second, analyzed RGB and HSV color space. RGB color space is selected for color image segmentation, according to advantages and disadvantages of the color space and the specific characteristics of image segmentation;Third, studied the M-PCNN model, expanded the traditional PCNN into a multi-channel PCNN, which is provided with multi-input. Specifically the internal activity of each channel is weighted coupling to get the overall internal activity. The occupied proportion of the pixel average value of each RGB component is as weighting coefficient, iterative threshold is modified as index increased dynamic threshold, color image segmentation is achieved by combining sync pulse payment feature with maximum entropy criterion. Experimental results show that, color image segmentation method based on M-PCNN, running time is reduced, detail of image segmentation is very clear.Finally, improved a three-dimensional pulse coupled neural network( 3D-PCNN). The two-dimensional connection coefficient matrix is expanded into three-dimensional matrix, PCNN is expanded from two-dimensional plane to three-dimensional space. In this paper, three-dimensional coefficient matrix of 26 neighborhood is convoluted with the corresponding color component, the maximum entropy criterion and automatic baud of neurons are combined to achieve color image segmentation, different areas of the image are displayed in different colors hierarchically based on 3D-PCNN model.