Effective representation using ICA for face recognition robust to local distortion and partial occlusion.

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Subspace projection methods in face recognition depend heavily on basis image characteristics.
• Robustness to local distortion and partial occlusion requires part-based local representation in basis images.

Purpose of the Study:

• To propose an effective part-based local representation method for face recognition.
• To enhance robustness against local distortion and partial occlusion using locally salient information.

Main Methods:

• Developed Locally Salient Independent Component Analysis (LS-ICA).
• LS-ICA imposes an additional localization constraint during Independent Component Analysis (ICA) basis image computation.
• Focused on utilizing locally salient information from important facial parts.

Main Results:

• LS-ICA demonstrated superior performance compared to PCA, ICA (Architecture I & II), LFA, and LNMF.
• The method showed particular effectiveness in scenarios with partial occlusions and local distortions.
• Experimental results validated the part-based local representation strategy.

Conclusions:

• LS-ICA offers a robust face recognition solution by leveraging part-based local representations.
• The method effectively addresses challenges posed by local distortions and partial occlusions in face images.
• Focusing on salient facial parts enhances the discriminative power of subspace projection methods.