Image feature extraction using the fusion features of BEMD and WCB-NNSC

Abstract

A novel image feature extraction method, using the fusion features obtained by the algorithms of Bidimensional Empirical Mode Decomposition (BEMD) and Weight Coding Based Non-negative Sparse Coding (WCB-NNSC), is proposed in this paper. This BEMD algorithm is on the basis of EMD, and is especially adaptive for non-linear and non-stationary 2D-data analysis. And the weight coding based NNSC algorithm includes more class information. Utilizing Intrinsic Mode Functions (IMF) extracted by BEMD to be the training set of the weight coding based NNSC algorithm, the feature basis vectors of natural images can be successfully learned, and these features behave locality, orientation, and spatial selection. Further, using extracted features, the image reconstruction task is implemented successfully. Moreover, compared with other feature extraction methods, such as FastICA, basic NNSC, WCB-NNSC and so on, simulation results show that our method proposed here is indeed efficient and effective in performing image feature extraction task. © 2011 Springer-Verlag.