Locality constrained joint dynamic sparse representation for local matching based face recognition

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Sparse Representation-based Classification (SRC) is widely used in face recognition.
• Variations in lighting, expression, and pose degrade SRC performance.

Purpose of the Study:

• To propose a robust face recognition method, Locality Constrained Joint Dynamic Sparse Representation-based Classification (LCJDSRC).
• To enhance face recognition accuracy despite variations in image conditions.

Main Methods:

• Face images are partitioned into smaller sub-images.
• Sub-images are sparsely represented using a novel locality constrained joint dynamic sparse representation algorithm.
• Representation results are aggregated for final recognition, treating local matching as a multi-task learning problem.

Main Results:

• The proposed LCJDSRC algorithm effectively accounts for latent relationships among sub-images.
• Locality information is incorporated into the representation process.
• Experiments on ORL, Extended YaleB, AR, and LFW databases demonstrate superior performance compared to state-of-the-art methods.

Conclusions:

• LCJDSRC offers a robust solution for face recognition challenges.
• The multi-task learning approach and consideration of locality enhance recognition accuracy.
• The method shows significant effectiveness across diverse benchmark datasets.