Dispersion Constraint Based Non-Negative Sparse Coding Neural Network Model

A dispersion constraint based non-negative sparse coding (DCB-NNSC) model is discussed in this paper. To ensure the sparsity in self-adaptive, the kurtosis criterion is used to measure the sparse priori knowledge of feature coefficients. And to enhance the capability of feature separability, the dispersion ratio of within-class and between-class of sparse coefficient vectors is utilized. Simulation results show that, just as those sparse coding (SC) and NNSC published, DCB-NNSC model can also simulate successfully the respective field of V1 in the primary visual system of human beings. Moreover, compared with common NNSC models, DCB-NNSC model behave clearer sparsity. Using DCB-NNSC features to test image reconstruction task, the results obtained prove further that the DCB-NNSC model is indeed effective in extracting image features and modeling the V1 mechanism of the primary visual system of human in application.