This paper proposes a novel Fisher Discrimination Dictionary Learning (FDDL) method for sparse representation based image classification. The FDDL aims to learn a structured dictionary where each sub-dictionary corresponds to a specific class label. The method improves pattern classification performance by incorporating the Fisher discrimination criterion into both the dictionary learning and coding coefficient optimization processes. The dictionary atoms are designed to have correspondence to class labels, enabling the reconstruction error after sparse coding to be used for classification. Additionally, the Fisher discrimination criterion is applied to the coding coefficients to ensure they have small within-class scatter and large between-class scatter. The proposed FDDL method is evaluated on benchmark image databases and shows superior performance compared to existing sparse representation and dictionary learning based classification methods. The FDDL method is applied to face, digit, and gender recognition tasks, achieving higher classification accuracy with a smaller dictionary size. The method is also shown to be effective in utilizing both the reconstruction error and sparse coding coefficients for classification. The FDDL method is compared with other state-of-the-art methods, demonstrating its competitive performance in various pattern recognition tasks. The paper also discusses the convexity of the FDDL objective function and presents the optimization procedures for the FDDL method. The experimental results show that FDDL outperforms existing methods in face recognition, digit recognition, and gender classification tasks.This paper proposes a novel Fisher Discrimination Dictionary Learning (FDDL) method for sparse representation based image classification. The FDDL aims to learn a structured dictionary where each sub-dictionary corresponds to a specific class label. The method improves pattern classification performance by incorporating the Fisher discrimination criterion into both the dictionary learning and coding coefficient optimization processes. The dictionary atoms are designed to have correspondence to class labels, enabling the reconstruction error after sparse coding to be used for classification. Additionally, the Fisher discrimination criterion is applied to the coding coefficients to ensure they have small within-class scatter and large between-class scatter. The proposed FDDL method is evaluated on benchmark image databases and shows superior performance compared to existing sparse representation and dictionary learning based classification methods. The FDDL method is applied to face, digit, and gender recognition tasks, achieving higher classification accuracy with a smaller dictionary size. The method is also shown to be effective in utilizing both the reconstruction error and sparse coding coefficients for classification. The FDDL method is compared with other state-of-the-art methods, demonstrating its competitive performance in various pattern recognition tasks. The paper also discusses the convexity of the FDDL objective function and presents the optimization procedures for the FDDL method. The experimental results show that FDDL outperforms existing methods in face recognition, digit recognition, and gender classification tasks.