Based On Pulse Coupled Neural Networks, Image Segmentation And Fusion Research

Pulse Coupled Neural Network (PCNN) is a new type of artificial neural network which appeared in 1990s.The neural model of PCNN mimics visual nerve cell activity , so it is based on the biology background. PCNN can be greatly applied to image processing because of its'unchanging characteristic on signal scale ,signal intensity ,signal distortion and rotation . Moreover, it also has the characteristic of translating 2-D variable to 1-D time sequences. These characteristics make it suitable for image processing environment .Therefore, PCNN gradually becomes an image processing method.This article did some study in image processing aspects ,such as image segment and pixle-level image fusion.For the image segment , we improved the prime model through the analysis of PCNN work mechanism , utilized the improved model in image segment and put forward an ending-rule. We also designed a self-adaptive and closed system based on the computer vision because of the network parameters'effect on result of image segment. The result of experiment proved the method is effective.We designed different algorithms to fuse different types of images according to PCNN's different characteristics through detailed analysis of PCNN. We picked up pixels on the local complemented images according to accumulation of pulses. We created two selection rules ,supplemented the existed methods and get a conclusion how to choose different rule on different types of images. We put forward a self-adaptive weighted image fusion method using PCNN's synchronous pulse mechanism. The direction of the weight vector depended on the local cross entropy. This method was applied to visual light image and infrared image fusion. On the different focused images ,we used wavelet to decompose images firstly, then we used weighted rule on the low-frequency component and PCNN to the high frequency-components. At last , we reconstructed an image combining the components .We compared different fusion results from different wavelet filters. The method can be easily adapted to different focused images fusion. The results of a large of experiments showed that effects of all algorithms mentioned above were very well.