Pulse Coupled Neural Network Image Segmentation And Image Retrieval

Pulse coupled neural network is the third generation artificial neural network which has strict biological background and is different from the traditional artificial neural network. Compared with traditional artificial neural network, PCNN has the characteristics of dynamic variable threshold, dual channel multiplication modulation characteristics, pulses issuing characteristics, simulating the fatigue and the refractory period of the biological neurons and so on. So it can better to simulate the realistic biological neurons and processing signal. Now, PCNN was also widely used in image processing field such as image segmentation, image fusion, pattern recognition, route optimization, image enhancement, edge detection, and image retrieval etc.This paper mainly analyzes and studies the original PCNN model and its basic theory, and through automatically determining the parameters of PCNN model to process image segmentation, thus parameter selection difficult problem of PCNN model is solved, using the simplified model of PCNN for image segmentation, and the information entropy time sequence (entropy sequence) as the image characteristic vector, reusing Euclidean distance to match with standard texture image in the image library, namely, calculate the similarity of detected images and the images of the library, image retrieval is proceeded. This makes the selective problem of the key parameters of PCNN model has been resolved, so PCNN is expediently applied to the image segmentation and image retrieval, its segmentation quality and retrieval results are good. The following is my research work and the research achievements in this paper:First, the paper summarized the original pulse coupled neural network model and the improved PCNN universal model.Second, the paper introduced PCNN application and principle in image processing.Third, the determined key parameters method of PCNN model is presented. Pulse coupled neural networks has important applications in image processing, but there are difficulties in choosing model parameters and over-smoothing problems of the image edges. In this paper, the image bilinear interpolation operator is performed, and the anisotropic diffusion characteristics in images are used to determine the link weight parameter of the model, by using genetic algorithms to solve the model parameters and the attenuation of the link strength threshold parameter, the automatic image segmentation is successfully implemented. Simulation results show the image segmentation by the algorithm embodies more image contour and edge details, so the algorithm possesses good computing performance.Fourth, using PCNN translation invariance, rotating invariance, scale invariance, local invariant features to extract image feature vector in image processing, and using Euclidean distance to match with standard texture image in the image library, the paper proposed the method of image retrieval based on PCNN. This method Combines adaptive segmentation with adaptive image feature extraction, and fully uses the image space location features and image gray characteristics, and the entropy sequence as image characteristics is applied to image retrieval. Experiments show that this method is effective.