Image Process Based On Adaptive PCNN Model

Recently, people carry out extensive and in-depth research on pulse coupled neural network(PCNN), and make it show a strong capacity in the field of image processing. PCNN is a kind of single layer neural network model. It does not need training to realize pattern recognition, image segmentation, and target classification, etc. Therefore it is very suitable for real-time image processing environment.As a kind of multiple parameters system, it may get the same network iteration characteristic for the same image with different sets of parameters. And the situation may be different when the image changes.So there is no standard for parameter setting, and ideal parameters should be obtained based on the experience.To obtain adaptive PCNN model, this paper combines the standard simplified PCNN model with PSO algorithm and artificial colony algorithm respectively, and realizes the adaptive segmentaion of images.As while the PCNN model is put forward based on mammalian visual neron model, which is more in line with the situation that human visual system is sensitive to the image edge detail information.This paper puts forward the image fusion algorithm that combines PSO-PCNN model with multi-scale decomposition. The high-frequency subband image represents the image edge detail. When chosing the high frequency coefficients of fused image, the gray value of the high-frequency subband pixels is directly taken as the external input of PCNN neurons. Low frequency part adopts the improved spatial frequency as incentives of PCNN, which can effectively suppress noise influence on the fused image and also can effectively choose the fusion coefficients of the fused image. So that it realizes the adaptive image fusion.Given that the fitness function of the traditional optimization algorithm is present in the logarithmic mechanism, this paper puts forward two new kinds of fitness function excluding the logarithmic mechanism, namely the product cross entropy and thr product image entropy, which can reduce the computional complexity and save the calculation time.