An Applied Research Of Pulse Coupling Neural Network For Image Processing

With the coming of digital era, image information has become the most important source for man to get information and the most vital means to make use of information because of its large information quantity, fast transmitting speed and wide application fields. The image processing technique has become to be an inevitable and robust tool in our scientific research, social production and our daily life. Image information is transmitted to the cerebral cortex via the continuous transmission of neuron signals in human brain visual system, while the cerebral cortex is the key part for visual system to process the image information. With its sound biological background, Pulse Coupled Neural Network(PCNN) has got its prominence in image processing.PCNN is based on the mammal neuron model built by Eckhorn after his research on synchronous vibration of the neuron pulse cluster in the cerebral cortex of cat. As the symbol of the third generation of Artificial Neural Networks, PCNN can perfectly imitate the human or mammal's visual system with its features of variable threshold, non-linear modulation and local synchronization of action potentials. It is a simplification and approximation of the real neurons, and can be used to various fields of image processing.The thesis conducts a research on the application of PCNN for the image processing. Firstly, the image processing context and common techniques is brief introduced such as image filtering and image segmentation. Secondly, the basic model and algorithm and action behavior for the PCNN are introduced. We focused on the mage filtering and image segmentation based on PCNN. Finally,it is the summary and conclusion. The future research direction of PCNN is also given at the end of the paper.Based on the image denoising characteristics of PCNN, the similar center value filter algorithm of PCNN is used for color image denoising. A simplified PCNN model is built based on the HIS color space. Simulation and experimental results show that the proposed algorithm can well protect the image detail and is better than the conventional algorithm for image denoising. Similarly, it is well suitable to the image segmentation due to PCNN's properties such as feature capturing, local synchronization of action potentials and automatic wave transmission. The thesis discusses the application of PCNN to image segmentation and proposes an improved method based on maximum entropy ciriterion to solve problems in the practice application. Experiment verifies that the new method has got better performance than the tradition method.