Sensitivity analysis of partitioning-based face recognition algorithms on occlusions

Abstract

Holistic Principal Component Analysis (PCA) and holistic Independent Component Analysis (ICA) methods require long training times and large storage spaces for the recognition of facial images. These drawbacks can be avoided by using partitioning-based methods, namely partitioned PCA (pPCA) and partitioned ICA (pICA), which yield similar performance for pPCA and improved performance for pICA method compared to the holistic counterparts of these methods for the recognition of frontal facial images. This paper demonstrates the sensitivity analysis of pPCA and pICA methods on several types and sizes of occlusions for the recognition of facial images with similar facial expressions. The recognition rates for pPCA and pICA over occlusions are contrary to the recognition rates of these methods on occlusion-free facial images with different facial expressions.