Biometric Sample Quality and its Application to Multimodal Authentication Systems

THIS THESIS IS FOCUSED ON the quality assessment of biometric signals and its application to multimodal biometric systems. Since the establishment of biometrics as an specific research area in late 90s, the biometric community has focused its efforts in the development of accurate recognition algorithms and nowadays, biometric recognition is a mature technology that is used in many applications. However, we can notice recent studies that demonstrate how performance of biometric systems is heavily affected by the quality of biometric signals. Quality measurement has emerged in the biometric systems on certain pathological samples. We first summarize the state-of-the-art in the biometric quality problem. We present the factors influencing biometric quality, which mainly have to do with four issues: the individual itself, the sensor used in the acquisition, the user-sensor interaction, and the system used for processing and recognition. After that, we give strategies to ensure the best possible quality of acquired biometric samples. Next, we present existing frameworks for evaluation of the performance of biometric quality measures. The relationship between human and automatic quality assessment, as well as the role of quality measures within biometric systems is then analyzed. Lastly, we summarize standardization efforts related to biometric quality and we point out further issues and challenges of the quality problem. The experimental part of the Thesis starts with the study of quality in fingerprint images. We evaluate the impact of selected image quality measures in the performance of the two most used approaches for fingerprint recognition using a multi-session and a multi-sensor database. It is observed high correlation between the different quality measures in most cases, although some differences are found depending on the sensor. The behavior of the two matchers under varying image quality conditions has been also found to be different. We then study the problem of quality assessment in off-line signature images. We present several measures aimed to predict the performance of off-line signature verification systems measuring factors like signature legibility, complexity, stability, duration, etc. We also present a new matcher based on local contour features, which is compared with two other approaches. Some remarkable findings of this chapter are that better performance is obtained with legible signatures and skilled forgeries, or that performance is worsened with highly variable signatures. Finally, we contribute with a quality-based multibiometric architecture that is generalizable to biometric systems working with multiple sources of information (different modalities, matchers, acquisition devices, etc.). In this approach, quality is used to switch between different system modules depending on the data source, and to consider only data of enough quality. We compare the proposed architecture with a set of simple fusion rules. It is demonstrated that the proposed system outperforms the rest when coping with signals originated from heterogeneous biometric sources, pointing out its effectiveness. An additional overall improvement of 25% is observed in the EER by incorporating a quality-based score rejection scheme, showing the benefits of incorporating quality information in biometric systems.