Search

In this paper, we tackle the twins identification problem; a challenging task in biometric authentication. The experiments are carried out using deep learning techniques and for that we have proposed a Convolutional Siamese Network (CNN-Siamese) and a multiscale Convolutional Siamese network (MCNN-Siamese). We have also explored and evaluated three pre-trained CNNs namely, ResNet-50, VGG-16 and NASNet-Large, and utilized them in Siamese network after appropriate modifications. In addition, a simple 5-layer neural network (sNN) is also utilised as Siamese subnetwork in the experiments. We have presented compelling experimental evidences in terms of correct classification accuracy (CCR) on CASIA-IrisV4 Twins’ dataset that manifest the effectiveness of ocular biometrics in identifying twins. The results achieve the existing state-of-the-art human level accuracy. Also, the proximity of CCRs of all the models asserts that the ocular regions in twins hold a significant correlation to label them as twins. We have also quantified the cross-domain capability of the proposed subnetworks (i.e. CNN and MCNN) on ND-GFI dataset that outperforms the state-of-the-art methods. Notably, for the ocular biometrics, to the best of our knowledge, there is currently no literature available as of date that explores the association between twins and subsequently unriddles the classification using Deep Learning.

The inner boundary of iris represents the pupil’s edge. Hence, to work an Iris Recognition System (IRS) and the gaze tracking system expeditiously it is important to locate it as precisely as possible in a significant amout of time. In the presence of non-ideal constraints e.g. non-uniform illumination, poor contrast, eyelashes, hairs, glasses, off-angle orientation, these systems may not work well. In this paper we present an adaptive pupil localization method based on the roundness criteria. First, it applies a gray level inversion to suppress the reflections, then it performs Gray level co-occurrence matrix (GLCM) based contrast estimation. If this estimated contrast is lower than a certain threshold, the input image is made to undergo gamma correction to adjust the contrast. Subsequently, anisotropic diffusion filtering followed by log transformation is applied, which suppresses the effect of eyelash occlusion, limits the creation of small regions and highlight the dark pixels. Afterwards, a clean binary image with few regions is acquired using adaptive thresholding and some morphological operations. Finally, the roundness metric is computed for each of these regions and the region with largest roundness metric, also being greater than a prescribed threshold, declared as pupil. Experiments were carried out on few well known databases, NICE1, CASIA V3 lamp, MMU, WVU and IITD. The results are grounded upon subjective and objective evaluation; which in turn, indicate that our method outperforms a state-of-the-art approach and a deep learning approach in terms of localization capability in some unconstrained scenarios and shorter processing time. After assessing the performance of the proposed algorithm, it is manifested that it ensures a fast and robust localization of pupil in the presence of corneal reflection, poor contrast, glasses and eyelash occlusion.

The main aim of this paper is to analyse the iris texture which becomes the basis of feature extraction. In Iris Recognition System (IRS), feature extraction is one of the significant steps which is followed by a matching process. The proposed method takes into account three techniques: DSIFT, HOG and DCT. These techniques are individually employed on the blocks that correspond to the regions of interest, not occluded by noise. A block-wise matching of obtained feature vectors is carried out using a standard measure. The true match of the system relies on the three aforementioned techniques which make the system more robust. The experimental results are reported on CASIA-IrisV1 database. The obtained outcomes indicate that the performance of the system increased significantly due to the score level fusion of the matching that makes it suitable for the iris recognition related applications.

A categorical study on iris localization, its challenges and techniques is carried out here. This survey comprehends the historical evolution and present state-of-the-art in iris localization. Most research publications include their primary contribution in iris localization based on tools like Hough Transform, Integro-differential Operator, Active Contour and Deep Learning etc. Majority of the techniques discussed in this paper are popular and efficient enough, however some are unique in their ways. Vast and growing horizon of iris biometric research forces this work to be selective. Our work will be of assistance to someone who is new to iris biometrics and wants to acquire a quick knowledge about iris localization, hurdles before it and various methods to achieve this.

This paper presents a multilevel image thresholding approach which relies on Tsallis entropy using Fuzzy partition with a novel threshold selection technique. In order to compute the optimal threshold values, Differential Evolution (DE) has been employed. The proposed method can further be exploited in image segmentation which is considered to be a critical step in image processing. Our proposed threshold selection technique is based on Tsallis-Fuzzy entropy and the results are compared with Shannon entropy (or fuzzy entropy) and Tsallis entropy based existing threshold selection techniques. The experiments are performed on two different sets of images and the results have been compared with that of existing state-of-the-art methods, namely, Patch Levy Bees' Algorithm (PLBA), Bacterial Foraging optimization (BFO), modified Bacterial Foraging optimization (MBFO) and Bees' Algorithm (BA). Quantitative analysis is carried out based on three image quality metrics viz SSIM, PSNR and SNR. Standard deviation and CPU time for convergence of the objective function have been calculated for performance evaluation. Furthermore, the statistical significance of our method has been estimated using Friedman test and Wilcoxon test. The experimental results manifest that our method produces results superior to the methods in comparison.

In this paper, image contrast enhancement is achieved by combining together the local–global image statistics and multiscale morphological filtering (MMF). The proposed method has been executed on two different sets of images, and the result has been compared with that of some existing standard methods, namely histogram equalization (HE), contrast limited adaptive histogram equalization (CLAHE), and multiscale morphology in order to have an outlook on the relative performances. The experimental results manifest that the proposed method produced results superior to the methods compared.