Multi-modality quality assessment for unconstrained biometric samples

The aim of this research is to investigate multi-modality biometric image quality assessment methods for unconstrained samples. Studies of biometrics noted the significance of sample quality for a recognition system or a comparison algorithm because the performance of the biometric system depends mainly on the quality of the sample images. The need to assess the quality of multi-modality biometric samples is increased with the requirement of a high accuracy multi-modality biometric systems. In this thesis we describe the work carried out to use no-reference image quality metrics for the assessment of multi-modality biometric sample images. The intended behavior of such metrics is to evaluate or predict the quality of natural image as human observers would perceive it. Following an introduction and background in biometrics and biometric sample quality, we introduce the concept of biometric sample quality assessment for multiple modalities. Recently established ISO/IEC quality standards for fingerprint, iris, and face are presented. In addition, sample quality assessment approaches which are designed specific for contact-based and contactless fingerprint, nearinfrared-based iris and visible wavelength iris, as well as face are surveyed. Following the survey, approaches for the performance evaluation of biometric sample quality assessment methods are also investigated. Based on the knowledge gathered from the biometric sample quality assessment challenges, we propose a common framework for the assessment of multi-modality biometric image quality. We review the previous classification of image-based quality attributes for a single biometric modality and investigate what are the common image-based attributes for multi-modality. Then we select and re-define the most important image-based quality attributes for the common framework. In order to link these quality attributes to the real biometric samples, we develop a new multi-modality biometric image quality database which has both high quality sample images and degraded images for contactless fingerprint, visible wavelength iris, and face modalities. The degradation types are based on the selected common image-based quality attributes. Another important aspect in the proposed common framework is the image quality metrics and their applications in biometrics. We first introduce and classify the existing image quality metrics and then conducted a brief survey of no-reference image quality metrics, which can be applied to biometric sample quality assessment. Plus, we investigate how no-reference image quality metrics have been used for the quality assessment for fingerprint, iris, and face biometric modalities. From the fingerprint quality ISO/IEC standard we already know what kind of degradations that can affect the performance of a fingerprint recognition system, however, it is not clear if the system performance will be influenced, or to what extent. Therefore, an experiment is conducted to investigate how a number of selected image degradations influence the performance of a fingerprint recognition system, and, influence the performance of a fingerprint quality assessment approach. According to the availability of open source biometric recognition applications, the existing contact-based fingerprint recognition algorithms have difficulties to extract the minutiae from the high quality contactless fingerprint images from our database. Therefore, we review 10 existing methods for the enhancement of contact-based fingerprint and select two suitable methods to verify whether they can be applied to contactless fingerprint. Finally, we propose an improved 3-step contactless fingerprint image enhancement approach for the minutia extraction. However, after the image enhancement process, the number of detected minutiae is still under an acceptable level. We decide to leave contactless fingerprint for our further study. The experiments for the performance evaluation of no-reference image quality metrics for visible wavelength face and iris modalities are conducted. The experimental results indicate that there are several no-reference image quality metrics that can assess the quality of both iris and face biometric samples. Lastly, we optimize the best metric by re-training it. The re-trained image quality metric can provide better recognition performance than the original. Through the work carried out in this thesis we have shown the applicability of no-reference image quality metrics for the assessment of unconstrained multi-modality biometric samples