Regression-Based Automated Facial Image Quality Model

Introduction

Despite the widespread use of automatic facial recognition systems, their consistent performance is affected when presented with uncooperative users and uncontrolled environments. Image quality has a very strong connection with the recognition performance and identification errors of a biometric system (Grother & Tabassi, 2007; Alonso-Fernandez et al., 2012; Bharadwaj et al., 2014). The quality of facial images during enrollment and verification stages significantly affects the performance of an automated face recognition system. Poor quality of data, due to changes in lighting conditions, facial expressions, pose, occlusion, and poor sensor quality, may cause efficiency loss. Recent studies show that variations in lighting conditions, resolution, camera movement, occlusion, expressions and other quality factors have a major impact on the recognition rate of a face biometric system (Sellahewa & Jassim, 2010; Abaza et al., 2012; Abaza et al., 2014; Punnappurath et al., 2015). Researchers have attempted to categorize the face-based quality factors based on digital formatting of facial images, scenes and photographics. Several techniques have been proposed in the literature to effectively compute different quality factors (Wang & Bovik, 2002; Abaza et al. 2012; Abaza et al., 2014; Sang et al. 2009). However, few attempts have been made to jointly consider various quality factors under one framework (Abaza et al., 2012; Abaza et al., 2014). Though researchers have considered a partial subset of these quality factors, there is yet no effective analysis of the impact of different quality factors on the performance of a face recognition system. Therefore, an effective quality estimation model is needed, which can characterize the quality of the facial image by integrating different quality measures into a single quality score which is an indicator of the overall quality of the facial image. The model should also reflect the impact of different quality factors on the face recognition performance.