[Paper Review] Examplers based image fusion features for face recognition

Alex Pappachen James, Sima Dimitrijev|arXiv (Cornell University)|Jan 28, 2012

Face and Expression Recognition21 references3 citations

TL;DR

This paper proposes an exampler-based image fusion method for face recognition that combines multiple gallery images of a person into discriminative features, using biologically inspired local binary decisions to improve accuracy and stability. The approach achieves 99.0–100.0% recognition accuracy across six benchmark datasets, outperforming single-model methods like face averaging.

ABSTRACT

Examplers of a face are formed from multiple gallery images of a person and are used in the process of classification of a test image. We incorporate such examplers in forming a biologically inspired local binary decisions on similarity based face recognition method. As opposed to single model approaches such as face averages the exampler based approach results in higher recognition accu- racies and stability. Using multiple training samples per person, the method shows the following recognition accuracies: 99.0% on AR, 99.5% on FERET, 99.5% on ORL, 99.3% on EYALE, 100.0% on YALE and 100.0% on CALTECH face databases. In addition to face recognition, the method also detects the natural variability in the face images which can find application in automatic tagging of face images.

Motivation & Objective

• To improve face recognition accuracy and stability beyond single-model approaches like face averaging.
• To model natural facial variability using multiple training samples per individual.
• To develop a biologically inspired method for local similarity decisions in face recognition.
• To enable automatic tagging of face images by detecting natural facial variations.
• To demonstrate superior performance on standard face recognition benchmarks using exampler fusion.

Proposed method

• Construct examplers by fusing multiple gallery images of the same person using a weighted fusion strategy.
• Apply local binary decisions based on similarity between test images and examplers at a pixel-wise or patch-wise level.
• Use a biologically inspired decision mechanism that mimics human visual perception of facial similarity.
• Integrate multiple training samples per identity to enhance robustness and reduce overfitting.
• Employ a voting or consensus mechanism over local decisions to determine final face classification.
• Utilize standard face databases (AR, FERET, ORL, EYALE, YALE, CALTECH) for evaluation under controlled conditions.

Experimental results

Research questions

• RQ1 Can fusing multiple gallery images per person improve face recognition accuracy compared to single-image models?
• RQ2 Does the exampler-based fusion method enhance stability and robustness in face recognition under varying conditions?
• RQ3 To what extent can the method detect and model natural facial variability for applications like automatic image tagging?
• RQ4 How does the biologically inspired local decision mechanism compare to conventional global feature extraction in face recognition?
• RQ5 What recognition performance can be achieved across diverse benchmark datasets using this exampler fusion approach?

Key findings

• The exampler-based method achieved 99.0% recognition accuracy on the AR database.
• The method reached 99.5% accuracy on both the FERET and ORL databases.
• It achieved 99.3% accuracy on the EYALE database and 100.0% on both the YALE and CALTECH databases.
• The fusion of multiple training samples per person significantly improved recognition performance over single-model baselines.
• The method effectively models natural facial variability, enabling potential applications in automatic face image tagging.
• The biologically inspired local decision mechanism contributed to higher stability and accuracy compared to conventional averaging techniques.