基于局部特征的非负稀疏编码神经网络模型

Non-negative Sparse Coding Neural Network Model Based on Localized Feature

在非负稀疏编码(NNSC)的基础上,考虑特征基向量的稀疏度约束和特征基的局部性,提出一种基于局部特征的NNSC神经网络模型。该模型利用梯度和倍增因子相结合的优化算法实现特征系数的学习;利用倍增算法实现特征基的学习。对掌纹图像进行特征提取测试,结果表明,与传统NNSC模型和局部非负矩阵分解(LNMF)方法相比,该模型能有效提取图像的局部特征,收敛速度较快,可模拟初级视觉系统处理自然界信息的稀疏编码策略。

On the basis of the Non-negative Sparse Coding（NNSC）,considered the sparse measure constraint of feature basis vectors and the locality of features,a novel NNSC model based on localized features is proposed in this paper.This NNSC model utilizes the optimized method that combines the gradient and multiplicative algorithm to learn the feature coefficients,and only the gradient algorithm to learn feature vectors.Using this NNSC model to test the feature extraction process of palm images,and compared with the NNSC model and Localized Non-negative Matrix Factorization（LNMF）,experimental results show that the model can extract image features efficiently and has quick convergence speed,as well as can model the sparse coding strategy used by the primary visual system in dealing with the nature processing.This further proves that the NNSC model proposed is feasibility and practicality in the theoretical research.