模拟视觉系统的非负稀疏编码神经网络模型  

Non-negative Sparse Coding Neural Network Models Based on Visual System

作　　者：尚丽[1] 苏品刚[1] 

机构地区：[1]苏州市职业大学电子信息工程学院,江苏苏州215104

出　　处：《苏州市职业大学学报》2014年第1期2-11,共10页Journal of Suzhou Vocational University

基　　金：国家自然科学基金资助项目(60970058)

摘　　要：非负稀疏编码(NNSC)神经网络模型能够有效模拟人脑初级视觉系统主视皮层V1区神经元的感受野,有效抽取自然界的特征,目前已在图像处理领域中得到广泛应用.考虑NNSC建模过程中稀疏先验分布的选取、特征基矩阵的稀疏度约束、特征基的最大化代表性、图像数据类别先验信息等主要因素,主要讨论了基于正态逆高斯(NIG)密度的双层反馈NNSC(NIG-N NSC)模型、基于局部特征的NNSC(LNNSC)模型以及基于Fisher线性判据的NNSC(FLD-NNSC)模型.研究结果表明,拓展的NNSC模型在图像特征提取、图像消噪和图像恢复中具有一定的实用性.Non-negative sparse coding neural network model can efficiently simulate the receptive field of neurons in the primary visual cortex V1 in primary visual system of brain and extract features of nature. Now this model has been used widely in the field of image processing. Considering some key influence factors,such as the selection of sparse prior distribution,the sparse constraint of feature matrix,the maximum representative-ness,class prior information of images and so on,several NNSC models are mainly discussed here including are models of NIG based NNSC with feedback mechanism denoted by NIG-NNSC,local-feature-based NNSC denoted by LNNSC,fisher-linear-discrimination-based NNSC denoted by FLD-NNSC,weighted-coding-based NNSC denoted by WCB-NNSC,etc. Research results testify that these extended NNSC models are applicable in the research field of image feature extraction,image denoising and image restoration.

关 键 词：非负稀疏编码 神经网络模型 稀疏分布 视觉系统 主视皮层V1区 特征基 图像处理 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]