Image Segmentation Based On Pulse Coupled Neural Networks And The Study Of Image Retrieval

Pulse coupled neural network PCNN (Pulse Coupled Neural Network) is the third generation of neural networks based on the traditional artificial neural network. PCNN evolves from cerebral cortex model Proposed by Eckhorn, which has translation, rotation, scaling, and distortion invariance in the image processing. PCNN has two characteristics which are different from the traditional model:Connect modulation and pulse capture. These characteristics make PCNN has been widely used in image processing. For example:image segmentation, image fusion, edge detection, pattern recognition and so on. In this paper, based on PCNN models and theoretical, two types problems are studied.First, PCNN applied in image segmentation is mainly based on its connection modulation, and similar clusters for the capture feature. Although the PCNN has achieved image segmentation, but there are a lot of segmentation noise points, and some edges are not well segmented. In response to this difficulty, this paper introduces bilateral filter algorithm, which has the role of edge-preserving and denoising. Experimental results show that the two algorithms can get the edges with more details, and get rid of a lot of noise points.Secondly, PCNN extracts features sequence with translation invariance, rotation invariance, scale invariance, distortion invariance and other features, but the actual images always be affected by uneven light, which brings a major problem to the image retrieval. To solve this problem, this paper introduces Gabor function, through the Gabor transform, the most energy of the image present in the DC part, so choosing high-frequency can avoid the lighting effects. After the Gabor processing, image basically removes the impact of light, PCNN can completely extract invariant image features. Experimental results show that the Gabor function and PCNN combined can achieve image retrieval.