A discriminative dictionary learning algorithm is put forward for face recognition.
The algorithm combines the linear classification error with the within-class scatter.
The fisher ratios of the coefficients are enhanced.
The proposed method outperforms some state-of-the-art dictionary learning algorithms.