Study on Image Recognition Based on PCNN and LVQ

Abstract

Pulse coupled neural network (PCNN) is the third generation of artificial neural network with strict biological background, which is different from traditional artificial neural networks. Compared with traditional feedback neural networks, pulse coupled neural network possesses significant characteristics from the point of composition of neuron. PCNN has the firing features of dynamic pulse, firing features of synchronization pulse, and dynamic and variable features of threshold value, which brings excellent effect of image segmentation and image denoising etc to PCNN. LVQ neural network is a competitive type neural network with supervised learning method; the algorithm is derived from Kohonen competitive algorithm. LVQ network is able to complete classification processing of extremely complicated patter recognition through interaction of internal elements only due to its simple structure, with excellent features for pattern recognition. This paper carries out study on the technology of image segmentation by virtue of PCNN; invariant moment features and textural features are extracted from different types of images to realize the effect of image dimension reduction. Finally, feature vectors will be sent to LVQ neural network to realize classification recognition. The experimental result indicates excellent segmentation effect and strong adaptability of this method, enabling better image recognition.