Assessment of fingerprint quality with application in enhancement and liveness detection

Abstract

Biometrics-based authentication has drawn attention of the researchers due to its widespread use in security and access control. Authentication of users can be done by utilizing various distinctive characteristics of the individuals such as: fingerprint, iris, face, etc. In particular, fingerprint-based authentication is the most widely adopted for personal identification due to its uniqueness and ease in sample acquisition. However, the performance of such systems relies on the quality of the acquired fingerprint images, as the low-quality fingerprint images degrade the recognition performance of the system. This performance degradation can be avoided if the low-quality fingerprint images are detected in the initial phase (fingerprint acquisition) of the fingerprint recognition system. Further, if the low-quality fingerprint images are acquired in the system, a quality adaptive fingerprint enhancement module can improve these low-quality images and avoid the performance degradation. The quality features of a fingerprint image are also utilized for detection of liveness in fingerprint images during the fingerprint acquisition. Therefore, the objective of this work is to enhance the performance of fingerprint recognition system through effective fingerprint quality analysis and quality adaptive fingerprint enhancement, and secure it from presentation attacks using quality features. In this dissertation, we have divided the problem of fingerprint quality assessment into two parts. First part comprises of fingerprint texture quality classification, in which the texture quality (dry, wet, and good) of fingerprint images is analyzed to remove the low quality fingerprint images. In the second part, a biometric quality score of fingerprint images is computed using various ridge-valley distribution based features. This method operates in two phases namely, block quality nature assessment and fingerprint quality score computation. In block quality nature assessment phase, different fingerprint blocks are assigned dry, normal dry, good, normal wet, and wet quality labels using ridge-valley distribution based features. In the second phase, based of the quality nature, orientation certainty analysis, and clarity analysis of each local minutiae patch, an overall quality score is assigned to the fingerprint image. In our next work, a quality adaptive fingerprint enhancement method is proposed. The proposed method works into two phases. In the first phase, a fingerprint quality assessment (FQA) method is designed to assign a suitable quality class to fingerprint images. In the second phase, two-stage fingerprint quality enhancement (FQE) is performed which com- prises of a quality adaptive preprocessing (QAP) method followed by either Gabor or STFT or ODF enhancement techniques. Finally, the effectiveness of quality features for fingerprint liveness detection is assessed in this thesis. In this work, we introduce a set of novel quality features for fingerprint liveness detection. The proposed method assesses the ridge-valley structure of live and fake fingerprint images and extracts relevant discriminative quality features. After feature selection, the selected feature set is fed to support vector machine (SVM), random forest (RF), and gradient boosted tree (GBT) classifiers to identify live or fake fingerprint images. The major contributions of the thesis are fingerprint texture quality classification, fingerprint quality assessment and scoring, quality adaptive fingerprint enhancement, and fingerprint liveness detection using quality features. Experiments evaluations performed on different databases confirm the potential robustness and superiority of the contribution made in this thesis.