Based On Pulse Coupled Neural Network Image Gaussian Noise And Mixed Noise Filtering

Pulse coupled neural network (PCNN) is the latest third generation of artificial neural network with biology visual characteristics published at Nature Journal in1989when GRAY discovered the phenomenon that the cat’s visual cortex neurons could stimulate the related concussion. The Gray’s model was modified by ECKHON based on the research phenomenon that the cat’s brain cortex could release pulses synchronously. Then the initial PCNN model was emerged.The PCNN model has strong processing capacity and adaptability by taking advantages of the coupling characteristics of neurons, combing with the transmission delay and bioelectricity attenuation characteristics and approving the biological characteristics of actual neural network caused by the neighboring neurons in nervous system of mammals producing pulses synchronously. Because the PCNN model do not need training, it is very suitable for processing images, such as filtering noise which is always the hot topic.In this thesis, a new Gaussian noise filtering method was produced by using the basic PCNN model. A novel mixed noise filtering algorithm was put forward based on the proposed PCNN model and the limited gray value. To reduce the influence of the parameters, a simplified PCNN model for filtering mixed noise was proposed. Research work and contributions of the dissertation are as follows:First, a novel method was brought forward for filtering Gaussian noise by using variable step time matrix of PCNN combing with kinds of filtering methods. Firstly, the time matrix of noised image was calculated based on the basic PCNN model. Subsequently, Gaussian noise was identified by the relationship between the central pixel and its neighboring pixels in a slipping window and then filtered via four methods: adding variable gray-scale, reducing variable gray-scale, wiener filter and median filter. And then the small noise was slipped by wiener filter. The experimental results showed that the proposed method had better filtering results in peak signal to noise ratio and improved signal to noise ratio compared with some other filtering methods.Second, a new method of filtering mixed noise based on limited gray-scale and pulse coupled neural network was proposed. Firstly the salt and pepper noise was identified according to limited gray-scale and then filtered by mean grays of non-noised pixels while reducing the effect of the noise and preserving image edges and details. Secondly, the Gaussian noise was located by using the time matrix of PCNN, and then it was filtered by adaptively selected filters based on variable step. The experimental results showed the new method had more advantages in terms of filtering effects, adaptiveness and preserving image details compared to other traditional processes.Third, a novel and effective mixed noise filtering method was proposed based on simplified pulse coupled neural network. The basic PCNN model was simplified in order to reduce the effect of various parameters. The mixed noise was filtered by the time matrix combined with the characters of salt and pepper noise based on the simplified PCNN model. Experimental results showed that the novel method had better filtering results and more advantages compared with some filtering methods based on the original PCNN model and several kinds of traditional methods.