Wavelet Transform-Based Feature Extraction for Face Kinship Verification
El Ouanas Belabbaci, Mohammed Khammari, Gharbi Akram Abderraouf, Berkani Mohamed Rafik Aymene, Chouchane Ammar · Preprints.org · 2024
Kinship verification using facial images has become an important area of research within computer vision. This paper presents a robust kinship verification system that leverages Histograms of a Two-Dimensional Discrete Wavelet Transform (Hist-2D-DWT) to improve recognition accuracy. The system extracts planar features by decomposing images into four coefficients—approximation, horizontal, vertical, and diagonal—via discrete wavelet transform. Comprehensive experiments conducted on the TSKinFace and Cornell KinFace datasets validate the method’s effectiveness, achieving accuracies of 93.46% and 89.61%, respectively.