Sparse coding for image classification base on spatial pyramid representation

Abstract

Many efforts have been devoted to apply sparse coding for image classification with the aim of minimizing the reconstruction error and classification error. So far, the approaches have been proposed either separate the reconstruction and classification process which leave rooms for further optimization or form a complicated training model which cannot be resolved efficiently. In this paper, we first propose extracting the spatial pyramid representation as the image feature which forms the foundation of dictionary learning and sparse coding. Then we develop a novel sparse coding model which can learn the dictionary and classifier simultaneously in which form we can get the optimal result and can be solved efficiently by K-SVD. Experiments show that the suggested approach, in terms of classification accuracy and computation time, outperforms other well-known approaches.