Optimized Gabor-Based Discriminant Locality Preserving Projection for Feature Extraction and Recognition

[ 31 May 2023 | vol. 16 | no. 1 | pp. 23-34 ]

About Authors:

Roman Neruda
-Academy of Sciences of the Czech Republic, Czech Republic

Abstract:

This paper introduces an innovative algorithm known as the Optimized Discriminant Locality Preserving Projections (ODLPP), which is based on Gabor features—a type of image processing method that mimics how humans perceive visual stimuli. The standout feature of this algorithm is its ability to optimize the discriminant locality preserving criterion directly within a rich, high-dimensional Gabor feature space. It achieves this through a sophisticated technique called simultaneous diagonalization, which allows for a more efficient analysis of the data. This advancement removes the requirement for preliminary dimensionality reduction, a common hurdle in traditional methods that often compromises the integrity of the data.

To demonstrate the effectiveness of the ODLPP algorithm, a comprehensive series of experiments were carried out using the VALID face database, a well-regarded repository for testing facial recognition systems. The results of these experiments indicate marked improvements in classification performance, emphasizing the algorithm's strength in enhancing feature representation while effectively maintaining crucial local structures within the data. These findings suggest that the ODLPP not only simplifies the overall processing pipeline—making it more efficient—but also achieves impressive and reliable results in face recognition tasks, thereby showcasing its potential for practical applications in various fields, including security and biometric identification.

Keywords:

Optimized Discriminant Locality Preserving Projection; Gabor Feature; Face Recognition; VALID face database

 

About this Article: