Subspace learning for face verification
Ching‐Ting Tu, Mengying Lin, Shih-Hsun Hsiao · 2017
Face verification has been widely studied due to its importance in surveillance and forensics applications. In this paper, our goal is to verify identity of a pairwies sample, where each sample contains two heterogeneous facial images captured under different scenarios (e.g., low-resolution vs. high-resolution, or occluded facial image vs. non-occluded one). In oder to address the appearance difference between these two images, a example-based verification framework is proposed. We adopt principal component analysis (PCA) approach to learn subspaces in order to capture the apperance correlation between pairwise images. Futhermore, a feature selection process is included to select eignvectors that maximize the identity discrimination between two images. Experiments on variety face verification tasks demonstrate the effectiveness of our proposed learning framework.