Enhanced Local Gradient Order Features and Discriminant Analysis for Face Recognition.

Area of Science:

• Pattern Analysis and Machine Intelligence
• Computer Vision
• Machine Learning

Background:

• Subspace learning with complex data is crucial in pattern analysis.
• Existing methods using single-type features like image gradient orientations (IGOs) struggle with complete variation characterization.
• Discontinuities in edge alignment and feature matching are inadequately addressed.

Purpose of the Study:

• To develop robust features for discriminant subspace learning in facial image analysis.
• To address limitations of traditional single-type features and improve handling of discontinuities.
• To enhance the intrinsic structure discovery of facial images.

Main Methods:

• Exploitation of local order constrained IGOs to generate robust features.
• Utilization of difference-based filters to enhance local textures and order-based coding.
• Automatic fusion of multimodal features in a discriminant subspace with adaptive interaction functions to suppress outliers.
• Modification of a sparsity-driven regression model for classification with compact feature representation.

Main Results:

• Proposed features enhance local texture characterization and order-based coding ability.
• Adaptive interaction functions effectively suppress outliers for robust similarity measurement.
• The modified sparsity-driven regression model adapts well to classification tasks.
• Experiments on benchmark datasets demonstrate the algorithm's effectiveness in controlled and uncontrolled environments.

Conclusions:

• Local order constrained IGOs provide a robust feature representation for complex data.
• The proposed method effectively fuses multimodal features and handles outliers for discriminant analysis.
• The algorithm shows promising performance in facial image classification, outperforming traditional methods.