The pivotal role of datasets in signature verification systems motivates researchers to collect signature samples. Distinct characteristics of Persian signature demand for richer and culture-dependent offline signature datasets. This study introduces a new and public Persian offline signature dataset, UTSig (University of Tehran Persian Signature), that consists of 8280 images from 115 classes. Each class has 27 genuine signatures, 3 opposite-hand signatures, and 42 skilled forgeries made by 6 forgers. Compared with the other public datasets, UTSig has more samples, more classes, and more forgers. The authors considered various variables including signing period, writing instrument, signature box size, and number of observable samples for forgers in the data collection procedure. By careful examination of main characteristics of offline signature datasets, they observe that Persian signatures have fewer numbers of branch points and end points. They propose and evaluate four different training and test setups for UTSig. Results of the authors’ experiments show that training genuine samples along with opposite-hand samples and random forgeries can improve the performance in terms of equal error rate and minimum cost of log likelihood ratio.

Challenges to contextual filtering techniques include difficulties in estimating orientation field in poor quality images and subsequently failure to extract ridges reliably in large regions of low quality. This study proposes a statistical-based and dynamic fingerprint preprocessing technique for adaptive contrast enhancement and binarisation of fair and poor qualities plain and rolled fingerprints with large regions of low quality, prior to orientation field estimation. The algorithm effectively enhances smudged and faded ridges uniformly in recoverable regions, based on values of statistical variables computed locally in each region. The preprocessing algorithm employs a locally adaptive thresholding approach resulting in enhanced binarised images. The performance of the proposed algorithm was determined by carrying out biometric verification evaluation using a popular commercial biometric matching software, on databases of fingerprints in their original forms, as well as same fingerprints enhanced with the proposed algorithm. Experiments show that fingerprints are uniformly enhanced and binarised; and smudged or faded ridges in recoverable regions made visible. Fingerprint verification evaluation on preprocessed fingerprints resulted in lower error rates in 12 databases. These results show that the proposed algorithm significantly improves recognition.

Biometric verification can be considered one of the most reliable approaches to person authentication. However, biometrics are highly sensitive personal data and any information leakage poses severe security and privacy risks. Biometric templates should hence be protected and impersonation with stolen templates must be prevented, while preserving system's performance. In this study, a general biometric template protection scheme based on honey templates and Bloom filters is proposed, in order to grant privacy protection to the enrolled subject and detect the use of stolen templates. The performance and security evaluations show the soundness of the proposed scheme for facial verification. The benchmark is conducted with the publicly available BioSecure Multimodal DB and the free Bob image processing toolbox, so that research is fully reproducible.

In this study, the authors propose a local descriptor based multimodal approach to improve face recognition performance. Pre-processing is done to smooth, resample, and register data. The resampled three-dimensional (3D) face data are applied to extract novel descriptors including pyramidal shape index, pyramidal curvedness, pyramidal mean, and Gaussian curvatures. Proposed pyramidal shape maps are extracted at each level of the Gaussian pyramid on each point of the 3D data to have 2D matrices as representatives of 3D geometry information. A local descriptor structural context histogram, which represents the structure of the image using scale invariant feature transform, is calculated on pyramidal shape map descriptors and texture image to find matched keypoints in 3D and 2D modality, respectively. Score-level fusion by means of sum rule is employed to get a final matching score. Experimental results on the Face Recognition Grand Challenge (FRGC v2) database illustrate verification rates of 99 and 98.65% at 0.1% false acceptance rate for all versus all and ROC III experiments, respectively. On Bosphorus database, verification rate of 95.8% for neutral versus all experiment has been achieved.

Hyperspectral imagery analysis has become a popular topic for improving face recognition accuracy. Nevertheless, it encounters difficulty in data acquisition, low signal-to-noise ratio, and high dimensionality. As a result, there exists a need to develop better algorithms in order to achieve higher classification rates. In this study, the authors propose a new method for hyperspectral face recognition with very competitive experimental results. Since there is a significant amount of noise in every spectral band, they reduce noise adaptively from each spectral band by using any image denoising method, e.g. block matching and 3D filtering. They then crop each face according to its eye coordinates so that translation invariance can be achieved. They conduct log-polar transform to each cropped face image and extract 2D Fourier spectrum from them. In this way, the extracted features are approximately invariant to translation, rotation, and scaling. They use the collaborative representation-based classifier with voting for hyperspectral face recognition. They perform some experiments to test the authors’ new method for hyperspectral face recognition with very promising results.