Two-dimensional maximum local variation based on image euclidean distance for face recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Manifold learning algorithms aim to improve high-dimensional data analysis, particularly for image classification.
• Existing methods often overlook the interplay between pixel relationships and image geometry in dimensionality reduction.

Purpose of the Study:

• To propose a novel linear approach for face recognition that addresses limitations in current manifold learning techniques.
• To enhance dimensionality reduction by incorporating both pixel-level image structure and global image properties.

Main Methods:

• Introduced two-dimensional maximum local variation (2DMLV) for face recognition.
• Utilized image Euclidean distance to capture pixel relationships, differing from conventional methods.
• Integrated local variation, reflecting image diversity and discriminative information, into the dimensionality reduction objective function.

Main Results:

• Extensive experiments validated the effectiveness of the proposed 2DMLV approach.
• The method demonstrated superior performance in face recognition tasks compared to existing techniques.

Conclusions:

• The 2DMLV method offers a significant advancement in face recognition by effectively leveraging image structure and geometry.
• This approach provides a robust framework for dimensionality reduction in high-dimensional image data.