Robust Face Recognition via Adaptive Sparse Representation (1404.4780v1)

Abstract: Sparse Representation (or coding) based Classification (SRC) has gained great success in face recognition in recent years. However, SRC emphasizes the sparsity too much and overlooks the correlation information which has been demonstrated to be critical in real-world face recognition problems. Besides, some work considers the correlation but overlooks the discriminative ability of sparsity. Different from these existing techniques, in this paper, we propose a framework called Adaptive Sparse Representation based Classification (ASRC) in which sparsity and correlation are jointly considered. Specifically, when the samples are of low correlation, ASRC selects the most discriminative samples for representation, like SRC; when the training samples are highly correlated, ASRC selects most of the correlated and discriminative samples for representation, rather than choosing some related samples randomly. In general, the representation model is adaptive to the correlation structure, which benefits from both $\ell_1$-norm and $\ell_2$-norm. Extensive experiments conducted on publicly available data sets verify the effectiveness and robustness of the proposed algorithm by comparing it with state-of-the-art methods.

• The paper introduces ASRC by integrating sparsity and correlation using l1 and l2 norms to adaptively select discriminative features.
• It demonstrates superior recognition accuracy through extensive experiments on Yale, ORL, and AR datasets under various perturbations.
• The study paves the way for future research in hybrid deep models and broader pattern recognition applications.

Robust Face Recognition via Adaptive Sparse Representation: An Overview

The paper "Robust Face Recognition via Adaptive Sparse Representation" presents a novel methodology for face recognition that integrates the principles of sparsity and correlation in the context of Sparse Representation-based Classification (SRC). SRC has previously achieved significant results in face recognition, primarily leveraging the sparsity of representations. However, the rigid emphasis on sparsity has led to inadvertent neglect of correlation information, which can be crucial, particularly in highly correlated data sets. The authors introduce Adaptive Sparse Representation-based Classification (ASRC) to address these limitations.

Key Contributions

1. Integration of Sparsity and Correlation: The core innovation of ASRC lies in its adaptive framework, balancing sparsity and correlation by employing both $\ell_1$-norm and $\ell_2$-norm. This dual consideration allows ASRC to adapt its representation strategy based on the correlation levels among the samples. For samples with low correlation, ASRC behaves similarly to SRC, selecting the most discriminative samples. In contrast, for high correlation samples, it leverages a broader subset of correlated samples, ensuring a robust representation.
2. Theoretical and Practical Validation: The paper provides an extensive theoretical analysis detailing how the proposed correlation adapter, formulated through trace norm regularization, adjusts the representation model to account for data structure effectively. The theoretical underpinning is complemented by empirical validation through experiments on well-known face recognition datasets such as Yale, ORL, and AR.
3. Superior Performance: ASRC demonstrates superior recognition accuracy compared to contemporary methods, including NN, NFS, SRC, CRC, and LSRC, across various datasets and experimental conditions, highlighting its robustness, especially against occlusions and other common face image perturbations.

Results and Implications

The results reported indicate that ASRC achieves significant performance improvements, especially noticeable in scenarios with high data correlation and limited training samples, where traditional SRC would typically struggle. ASRC’s adaptability to data structures implies a less volatile performance degradation in face of extreme conditions, such as significant data occlusion—situations that are often encountered in real-world applications.

By effectively balancing between sparsity and correlation, ASRC showcases an enhanced capability for managing variations caused by different poses, expressions, and illumination conditions—critical factors in face recognition applications. These findings suggest a broader applicability of ASRC beyond face recognition into other fields requiring robust pattern recognition capabilities, such as activity recognition and motion segmentation.

Future Directions

The adaptive framework proposed by ASRC opens up several avenues for future research. Exploring its applications in domains outside of face recognition could provide insights into its generalizability and potential customization for other tasks. Furthermore, integrating ASRC with emerging models like deep learning could yield powerful hybrid approaches that leverage deep architectures' feature extraction capabilities alongside ASRC's adaptive representation strengths. Investigating these hybrid approaches represents a promising direction for enhancing pattern recognition systems' robustness and efficacy.

In conclusion, the contributions of this paper significantly enhance the capabilities of SRC-based methodologies. By addressing both sparsity and correlation jointly, ASRC offers an advanced approach to face recognition that can outperform existing techniques and adapt effectively to complex, real-world data challenges.