A Method Of Image Mixed-noise Removal Base On Modified PCNN Model

In the process of image forming and transmission, image is inevitably polluted by noise which will have an adverse impact on image visual effect and subsequent processing, even directly affect the follow-up image processing results. In actual applications, image usually is polluted by mixture-noise that is composed of different kinds of noise, rather than a kind of noise. So mixed-noise filtering of image is of great significance.Pulse coupled neural network (PCNN) is a new kind of neural network model proposed recently by simulating the optic neuron of animal cerebral cortex. Because of its unique biological background and characteristics, has successfully been applied to image filtering, image segmentation, feature extraction, object recognition etc. As PCNN model is simple to apply to image processing and easy to use integrated circuits to implement, so that it is possible to real-time image processing. Therefore, the research of method of mixed noise filtering base on PCNN is a very interesting research topic.This paper firstly elaborates on the basic theory and operating mechanism of PCNN, analyzes and summarizes its structure features and characteristics. Simplified and modified PCNN model appropriately on the foundation of referring existent theory and ideas, it is analyzed in detail the operating mechanism and parameter settings of the modified model.Secondly, this paper analyzes the characteristic of mixed noise constituted by pulse noise and Gauss noise and denoises mixed-noise in gray-scale image using a modified PCNN model combined with median filtering. Generally, we need to manually choose an appropriate network model parameter to achieve an optimal filtering result. In order to eliminate this trivial task, this paper proposes automatic mechanism to adaptively generate an optimal value for the network model parameter based on our modified model. Simulation results demonstrate that the denoising performance of our automatic mechanism is better than conventional algorithms that using manually parameter choosing. Meanwhile, our method is able to preserve the details in images.Also, this paper compares the proposed modified PCNN filtering algorithm to median filtering and fuzzy rule based filtering algorithm. The results show that the filtering performance of our algorithm is better than the others. In particular, our algorithm can achieve a more obvious improvement as the mixed-noise level of the images is getting higher.Finally, concerning mixed-noise denoising in color images, this work proposes a mixture noise filtering approach for color images in RGB color space based on the modified PCNN model. A large number of experiments show that the modified PCNN filtering algorithm can denoise the mixed-noise in color images effectively.