The Study And Application Of Image Segmentation Based On Pulse-Coupled Neural Network

Pulse coupled neural network (PCNN)is a new neural network, which was proposed based on phenomena regarding the neuron spiking synchronization of the animals visual cortex about observed characters in the experiment. PCNN directly comes from researches about animal visual property and has a property causing adjacent neurons with the comparability feature to emit synchronously pulse. This will recuperate spatial incoherence and little change in the extent about imported datum, so that region information in the image will be mostly hold and those will make for image segmentation. The properties have not only important theoretical significance, but have a widely future in the application.The magnitude of image information may be reflected by the entropy which acts as statistic peculiarity in image analysis. As for image segmentation, the entropy value is larger, the better image information from original image and the more details of the image are included, and the segmentation effect is better. By using the fuzzy entropy rule which is used to estimate fuzzy extent of the segmented image, the segmented image will include the most contents from original image and optimal segmentation effect will be obtained. The segmented image which has larger fuzzy entropy is the optimal result in the series of segmented image.Firstly, the research shows that PCNN model and simple PCNN model are used to different typical images and medical images. Compared with the Otsu method, researches demonstrated that the two PCNN models can be applied to image segmentation efficiently. Appling the two models to the image segmentation, we obtained better results. The experimental results demonstrated the better validity and robustness of the two models.By combining the conception of fuzzy entropy, then this paper improves the traditional PCNN model according to the maximal fuzzy entropy principle and presents automatic segmentation algorithm based on the maximal fuzzy entropy principle. The proposed method makes not only the most contents from the original image included in the segmented image and the optimal segmentation effect obtained, but the number of iterations will be automatically confirmed, so that the automatic segmentation about PCNN will be realized. There is a realistic significance in the practical applications.