Image Recognition Using Local Features Based NNSC Model

Non-negative sparse coding (NNSC) model can extract efficiently image features and has been used widely in image processing. However, in terms of the image feature classification, the recognition precision of NNSC is not ideal. To overcome this defect, on the basis of the basic NNSC model, considered the constraints of the sparse measure of feature basis vectors and the local features, a locality based NNSC (LNNSC) model is proposed here. In this NNSC model, the feature coefficients are learned by the optimized method that combines the gradient and multiplicative factor, and the feature basis vectors are learned by only the gradient algorithm. Selected palmprint images from PolyU database and considered different dimensions of image features, the results of feature extraction and recognition obtained by LNNSC are discussed. Furthermore, compared with other feature extraction methods, experimental results show that our NNSC 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 also proves that the LNNSC model is feasibility and practicality in the theoretical research.