Two No-Reference Image Quality Assessment methods based on Possibilistic Choquet Integral and Entropy: Application to Automatic Fingerprint Identification Systems

Quickly assessing image quality and rejecting poor-quality images can substantially improve current biometric systems. The main idea behind our proposed context-based Image Quality Assessment (IQA) approaches is to offer an alternative to learning by using contextual information (domain knowledge). This is achieved by using a set of Contextual Quality Indicators (CQIs) and transforming them into Quality Assessment Criteria (QAC). Proper transforming ways are via the Possibilistic Choquet Integral (PCI) and Possibilistic Entropy (PE). The choice of the CQIs to be used is based on a priori analysis of the context of application. We use ‘fingerprint’ as the context of application. The performance of an Automatic Fingerprint Identification System (AFIS) is impacted by poor-quality fingerprint images. The main advantage of our context-based approaches over other IQA methods is that it requires no learning stage, use only four statistical CQIs, and can act very early in the processing chain (i.e., on raw images) to reject quickly poor-quality images so that the subsequent AFIS can process only good-quality images, which result in better recognition rate (RR) performance. The functioning of two methods (PCI, PE) is illustrated using image samples from PolyUHRF database. The performance of over two proposed NR-IQA (PCI, PE) is compared with two well known learning-based NR-IQA methods (NIQE, SSEQ) and the well known Fingerprint-IQA (F-IQA) methods (NFIQ, RPS, OCL, GaborShen, Gabor, RVU and FDA). Four experimental Data sets, FVC2000, FVC2002, FVC2004 and FVC2006, have been used to make the comparison. The results obtained with the context-based approaches compare favorably with those obtained with learning-based methods for fingerprint domain of applications.