Iterative Re-Constrained Group Sparse Face Recognition With Adaptive Weights Learning.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Robust face recognition is crucial for security and identification systems.
• Existing methods struggle with variations like occlusion, illumination changes, and corruption.
• Sparse representation-based classification offers a promising avenue for handling such variations.

Purpose of the Study:

• To develop a robust face recognition method that addresses challenges like occlusion, corruption, and illumination variations.
• To propose an Iterative Re-constrained Group Sparse Classifier (IRGSC) with adaptive weights learning.
• To enhance discriminative power and structural information encoding in face recognition.

Main Methods:

• Introduced a Group Sparse Representation Classification (GSRC) approach with weighted features and groups.
• Developed an efficient algorithm to optimize the proposed objective function.
• Incorporated adaptively learned weights and locality structure into l2,p-norm regularization for a unified formulation.

Main Results:

• The proposed IRGSC method demonstrated significant improvements in performance and efficiency.
• Experiments showed superior robustness against face occlusion, corruption, and illumination changes compared to state-of-the-art methods.
• The derived algorithm's convergence was theoretically proven, ensuring a stationary point solution.

Conclusions:

• IRGSC is a robust and discriminative classifier for face recognition.
• Adaptive weights learning and group sparsity effectively enhance recognition accuracy under adverse conditions.
• The method's flexibility with training set size and feature dimensions makes it broadly applicable.