Region Selection for Robust Face Verification using UMACE Filters

Abstract

In this paper, we investigate the verification performances of four subdivided face images with varying expressions. The objective of this study is to evaluate which part of the face image is more tolerant to facial expression and still retains its personal characteristics due to the variations of the image. The Unconstrained Minimum Average Correlation Energy (UMACE) filter is implemented to perform the verification process because of its advantages such as shift–invariance, ability to trade-off between discrimination and distortion tolerance, e.g. variations in pose, illumination and facial expression. The database obtained from the facial expression database of Advanced Multimedia Processing (AMP) Lab at CMU is used in this study. Four equal sizes of face regions i.e. bottom, top, left and right halves are used for the purpose of this study. The results show that the bottom half of the face region gives the best performance in terms of the PSR values with zero false accepted rate (FAR) and zero false rejection rate (FRR) compared to the other three regions.