Supervised dictionary learning using distance dependent indian buffet process

This paper proposes a novel Dictionary Learning (DL) algorithm for pattern classification tasks. Based on the Distance Dependent Indian Buffet Process (DDIBP) model, a shared dictionary for signals belonging to different classes is learned so that the learned sparse codes are highly discriminative which can improve the pattern classification performance. Moreover, using this non-parametric method, an appropriate dictionary size can be inferred. The proposed method evaluated on different standard databases demonstrates higher classification accuracy than other existing DL based classification methods.