Evaluation of Feature Extraction Algorithms for Face Recognition Using ISO/IEC 25010

Abstract

Face recognition combines feature extraction, pattern comparison, and decision criteria, but accuracy alone does not fully describe implementation quality. The study aimed to comparatively evaluate EigenFace, FisherFace, and Viola-Jones for facial feature extraction using an ISO/IEC 25010-based framework. An experimental comparative study was conducted with the Yale Face dataset, split into training and test data. The three algorithms were implemented in Python within an Anaconda/Jupyter environment, using PCA as a common comparison basis and OpenCV for image processing. Functional suitability, performance efficiency, usability, reliability, maintainability, and portability were assessed through indicators scored from 1 to 5. Viola-Jones achieved the highest recognition effectiveness (95.0%), followed by FisherFace (90.0%) and EigenFace (86.6%); however, its execution took longer (25 min versus 9 and 7 min). In overall software-quality scoring, FisherFace reached 56/70, EigenFace 54/70, and Viola-Jones 53/70. The findings indicate that algorithm selection should not rely solely on accuracy because functionality, resource utilization, maintainability, and portability affect the technical decision. FisherFace achieved the most favorable overall balance in the controlled environment, whereas Viola-Jones stood out for functional accuracy.