Transparent contact lens spoofing has been demonstrated to hamper the overall performance of an iris recognition system. Achieving high detection accuracy with transparent lens is quite challenging using iris texture analysis based techniques. In this regard, the authors propose a supervised learning based transparent lens detection method by designing a novel set of features to describe the faint lens boundary. The input image is first segmented for extracting salient edge points in the sclera region of interest. RANSAC algorithm then performs circle fitting to generate the set of R-points for feature computation. The novel features such as concentricity parameter, R-point count, R-point intensity, R-circle radius, R-cluster count and circle genuineness are associated with distinctive range of values for each class. Experimental results obtained using the kernel-support vector machine (SVM) classifier show that the proposed method can achieve higher average detection accuracy as compared to state-of-the-art techniques: 90.63% (NotreDame I), 84.5% (NotreDame II), 83% (IIIT-D Cogent) and 86.27% (IIIT-D Vista).

The discriminating powers of biometric patterns derive from their entropy, just as the hardness of cryptographic keys derive from their entropy. The larger the number of independent bits, or the more independent they are, the less chance of collision. The authors measured the mutual information entailed by radial correlations within each of 632,500 different iris patterns from persons of 152 nationalities. For each iris, they measured how well the sequence of bits in any ring of the IrisCode predicts the sequence of bits in the other rings. Information density is quite non-uniformly distributed across iris patterns radially. Their measurements of mutual information address how much radial resolution is productive to use when encoding an iris, and they show that a non-uniform allocation of encoding resolution radially leads to significant performance improvements by reducing redundancy.

There is a noticeable tendency to apply deep convolutional neural network (CNN) in facial identification, since it is able to boost performance in face recognition and verification. However, due to the users have unique facial, exposure of face template to adversaries can severely compromise system security and users’ privacy. Here, the authors propose a face template protection technique by using multi-label learning, which maps the facials into low-density parity-check (LDPC) codes. Firstly, a random binary sequence is generated to represent a user and further hashed to produce the protected template. During the training, the random binary sequences are encoded by an LDPC encoder to produce diverse binary codes. Based on carefully designed deep multi-label learning, the facial features of each user are mapped to a diverse binary code. In the process of recognition and verification, the deep CNN mapping architecture is modelled as a Gaussian channel, while the noise brought by intra-variations in the outputs of CNN can be removed by the LDPC decoder. Thus, a robust face template protection scheme is achieved. The simulation results on PIE and extended Yale B indicate that the proposed scheme achieves high genuine accept rate at 1% false accept rate.

Face recognition is confronted with situations wherein the captured face images are not as clear as the registered ones; this is known as low-resolution face recognition. To solve this problem, the authors propose a new sparse coupled projection method using multidimensional scaling joint L _2,1-norm regularisation (MDSL21). The MDSL21 maps the low-resolution faces and high-resolution faces into a common sparse subspace in which feature selection and coupled transformation can be simultaneously achieved. In their proposed method, the authors first learn the common responds using the multidimensional scaling model. Specifically, the distance between the responds is approximated to the distance between the high-resolution samples, and the local manifolds are preserved. Then, the seeking coupled projections are formulated as a regression model. Inspired by the sparse constraint utilised in the classical subspace learning methods, the authors add an L _2,1-norm regularisation term to the regression model to realise the sparsity and present its optimisation method. Experimental results validate the effectiveness of their proposed method on the low-resolution face recognition task.

Palm vein recognition is motivated by the advantages of high security and liveness detection, but its popularity is prevented by the cost of palm vein capture devices. This study proposes a low-cost and practical palm vein recognition system. First, the authors’ system captures near-infrared (NIR) palm vein image with complementary metal–oxide–semiconductor camera in lieu of an NIR charge-coupled device camera. The goal is to reduce the cost of palm vein capture devices greatly. Second, this study adopts thenar area on the palm as the region of interest (ROI) for further palm vein recognition. The goal is to get the rich vessel and avoid the effect of palmprint. Finally, the discriminate palm vein features are extracted based on Haar-wavelet decomposition and partial least squares algorithm on the ROI image. The goal is to increase the recognition accuracy, though the resolution of the image is low. A database with 1500 palm vein images from 250 samples is setup with the capture device. Experiments in the self-built database and a public database show the effectiveness of the scheme.

Reduced training sets are major problems typically found on the task of offline signature verification. To increase the number of samples, the use of synthetic signatures can be taken into account. In this work, a new method for the generation of synthetic offline signatures by using dynamic and static (real) ones is presented. The synthesis is here faced under the perspective of supervised training: the learning model is trained to perform the task of online-to-offline signature conversion. The approach is based on a deep convolutional neural network. The main goal is to enlarge offline training dataset in order to improve performance of the offline signature verification systems. For this purpose, a machine-oriented evaluation on the BiosecurID signature dataset is carried out. The use of synthetic samples (in the training phase) generated with the proposed method on a state-of-the-art classification system exhibits performance similar to those obtained using real signatures; moreover, the combination of real and synthetic signatures in the training set is also able to show improvements of the equal error rate.

This paper investigates the problem of writer identification from handwriting samples in Arabic. The proposed technique relies on extracting small fragments of writing which are characterised using two textural descriptors, Histogram of Oriented Gradients (HOG) and Gray Level Run Length (GLRL) Matrices. Similarity scores realised using HOG and GLRL features are combined using a number of fusion rules. The system is evaluated on three well-known Arabic handwriting databases, the IFN/ENIT database with 411 writers, the KHATT database with 1000 writers, and QUWI database with 1,017 writers. Fusion using the ‘sum’ rule reports the highest identification rates reading 96.86, 85.40, and 76.27% on IFN/ENIT, KHATT, and QUWI databases, respectively. The results realised on the KHATT database are comparable to the state of the art while those reported on the IFN/ENIT and QUWI databases are the highest to the best of authors' knowledge.