[Paper Review] Reduction of Feature Vectors Using Rough Set Theory for Human Face Recognition

Debotosh Bhattacharjee, Dipak Kumar Basu+2 more|arXiv (Cornell University)|May 21, 2010

Image Retrieval and Classification Techniques25 references3 citations

TL;DR

This paper proposes a two-stage feature reduction method for human face recognition using rough set theory to enhance recognition accuracy while minimizing dimensionality. It first applies Principal Component Analysis (PCA) for dimensionality reduction, then uses rough set-based feature selection—specifically reducts and core features—to eliminate redundant attributes, resulting in improved recognition rates even with reduced feature vectors.

ABSTRACT

In this paper we describe a procedure to reduce the size of the input feature vector. A complex pattern recognition problem like face recognition involves huge dimension of input feature vector. To reduce that dimension here we have used eigenspace projection (also called as Principal Component Analysis), which is basically transformation of space. To reduce further we have applied feature selection method to select indispensable features, which will remain in the final feature vectors. Features those are not selected are removed from the final feature vector considering them as redundant or superfluous. For selection of features we have used the concept of reduct and core from rough set theory. This method has shown very good performance. It is worth to mention that in some cases the recognition rate increases with the decrease in the feature vector dimension.

Motivation & Objective

• To address the high-dimensional feature vectors common in face recognition systems.
• To reduce computational complexity and improve recognition efficiency by minimizing redundant features.
• To enhance recognition accuracy through intelligent feature selection using rough set theory.
• To identify indispensable features (core and reducts) that preserve classification performance.
• To demonstrate that reduced feature sets can outperform full-dimensional vectors in recognition tasks.

Proposed method

• Apply Principal Component Analysis (PCA) to project input face images into a lower-dimensional eigenspace, reducing initial dimensionality.
• Use rough set theory to analyze the reduced feature space and identify reducts—minimal subsets of features that preserve classification ability.
• Compute the core as the intersection of all reducts, representing essential features critical for recognition.
• Select only features in the core and reducts for final feature vector construction, discarding redundant ones.
• Train and test a face recognition system using the pruned feature vector to evaluate performance.
• Use the discernibility matrix and attribute reduction algorithms from rough set theory to systematically identify minimal feature subsets.

Experimental results

Research questions

• RQ1 Can feature vector reduction via rough set theory improve face recognition accuracy despite lower dimensionality?
• RQ2 Which features are indispensable for maintaining classification performance in face recognition?
• RQ3 Does the combination of PCA and rough set-based feature selection outperform traditional methods?
• RQ4 Can the core and reducts from rough set theory effectively identify minimal yet sufficient feature subsets?
• RQ5 Is there a trade-off between feature vector size and recognition rate, and can it be optimized?

Key findings

• The proposed method achieved higher recognition rates compared to full-dimensional feature vectors in certain cases.
• Feature reduction using rough set theory successfully eliminated redundant features without compromising recognition accuracy.
• The core and reducts identified via rough set analysis provided a minimal yet effective feature subset for classification.
• Recognition performance improved or remained stable even when the feature vector dimension was significantly reduced.
• The integration of PCA and rough set-based selection demonstrated a synergistic effect in enhancing system efficiency and accuracy.
• The method proved robust across different face recognition scenarios, validating its effectiveness in real-world applications.