SpikeNet的研究及其在快速人脸识别中的应用  

Research on SpikeNet and Its Application in Quick Face Recognition

作　　者：张锦娟[1] 师军[1] 于佳丽[1] 卢照[1] 

机构地区：[1]陕西师范大学计算机科学学院,陕西西安710062

出　　处：《计算机技术与发展》2010年第7期235-239,共5页Computer Technology and Development

基　　金：陕西省自然科学基金资助项目(2005A12)

摘　　要：对SpikeNet网络的研究,其主要目的在于根据电生理学的研究成果,通过计算机构造与人脑相似的神经元及其网络拓扑结构来模拟人脑的思维过程,进而学习人脑对外界事物的反应过程,实现对人脸图像中的兴趣区域识别和智能处理。对SpikeNet的研究将有助于提高网络的识别速度和识别效率,对人工智能的研究和发展具有深远的意义。SpikeNet的理论基础是基于激发(spike)原理的排序编码(rank code ordering)研究思想。在数以万计的视觉神经元中,大脑可以将神经元按照敏感性排序,对于特定情况敏感的物体,特定的神经元会及早突破阈值而激发产生Spike,并通过多层网络不断传递信息。对人脸识别的实验表明,SpikeNet在速度上达到实时的神经计算,并且能很好地克服对比度和噪声的影响,获得比较理想的实验效果。The significance of the research on SpikeNet is to construct the neural network which could simulate the signal processing in brain neural network through the computer.This could help us with image processing on learning how the brain reflects the environment in visual field.The research could help to improve both the processing speed and the rate of image recognition.The basic idea behind SpikeNet is rank code ordering that there are billions of neurons in the visual system,brain could use the order in which neurons fire spikes to code information with different sensitivity.For the sensitive neurons,they can fire a spike first which present the different order of spikes and with these spikes,network could pass the information on and on.The face testing with different contrast and luminance values shows that the result of the experiment indicates that our system is feasible.

关 键 词：机器视觉 SpikeNet 排序编码 前馈式神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]