Application And Study Of Image Segmentation Using Pulse Coupled Neural Network

Pulse Coupled Neural Network (PCNN), which has the biology background, is a new generation artificial neural network .It is largely different from traditional artificial neural network. It shows the character of pulse emission, double-channel-multiplying modulation, can vary the threshold dynamically, has the refractory period and capturing period when the pulses coming from external neural cells and the external stimulation exist, shows the character of space-time accumulation, simulates well the fatigue, refractory period, pulse stimulation for biology neural cells. Now, PCNN has been used widely in image processing, and has been shown its enormous advantages. Image segmentation is primary step in image processing, which decide the precision of next processing. So these have been hot spots of study.The dissertation did further research on the image segmentation based on PCNN: Analysis of a variety of image segmentation techniques based on PCNN, for example, regional growth, image entropy, maximal correlative criterion and genetic algorithms, etc. Propose future research of image segmentation based on PCNN. Using improved PCNN model and maximum correlation criterion for image segmentation.Research work and new contributions of the dissertation are as follows:1. The dissertation summarized the PCNN model and principle, normal improved model and principle, Unit-Linking PCNN model and principle.2. Analysis image segmentation methods based on PCNN by way of summary. For example: PCNN image segmentation based on image entropy method, PCNN image segmentation based on maximal correlative criterion method, etc. Analysis of their specific application in image segmentation, propose future research of image segmentation based on PCNN.3. There are many parameters in the original PCNN model, and the process of parameters selecting is too complex, Unit-Linking PCNN model is simple, but it weaken similar nerve-cell's pulse emission. we use improved PCNN model to solve the above problem well in this paper. we also discuss parameters' influence to segmentation in normal improved model.4. For the treatment of thresholdθj ( n), previous method is to traverse by iterative decreasing. The two-dimensional maximum correlative criterion for solving the fixed threshold, eliminating the need for iterative decreasing step of threshold, greatly reducing the run time.5. The dissertation brought forward using maximal correlative criterion in segmentation iteration. It avoids large logarithmic computation in minimum cross-entropy. It also avoids the disadvantage of that distribution of pixel gray level and is irrelevant between segmented image and the original one.