Abhishek Kumar Gangwar

DeepIrisNet: Deep iris representation with applications in iris recognition and cross-sensor iris recognition

Despite significant advances in iris recognition (IR), the efficient and robust IR at scale and in non-ideal conditions presents serious performance issues and is still ongoing research topic. Deep Convolution Neural Networks (DCNN) are powerful visual models that have reported state-of-the-art performance in several domains. In this paper, we propose deep learning based method termed as DeepIrisNet for iris representation. The proposed approach bases on very deep architecture and various tricks from recent successful CNNs. Experimental analysis reveal that proposed DeepIrisNet can model the micro-structures of iris very effectively and provides robust, discriminative, compact, and very easy-to-implement iris representation that obtains state-of-the-art accuracy. Furthermore, we evaluate our iris representation for cross-sensor IR. The experimental results demonstrate that DeepIrisNet models obtain a significant improvement in cross-sensor recognition accuracy too.

Person recognition based on fusion of iris and periocular biometrics

In this paper we proposed a novel multimodal biometric approach using iris and periocular biometrics to improve the performance of iris recognition in case of non-ideal iris images. Though iris recognition has the highest accuracy among all the available biometrics, still the noises at the image acquisition stage degrade the recognition accuracy. The periocular region can act as a supporting biometric, in case the iris is obstructed by several noises. The periocular region is the part of the face immediately surrounding the eye. The approach is based on fusion of features of iris and periocular region. The approach has shown significant improvement in the performance of iris recognition. The evaluation was done on a test database created from the images of UBIRIS V2 and CASIA iris interval database. We achieved identification accuracy upto 96 % on the test database.

IrisSeg: A fast and robust iris segmentation framework for non-ideal iris images

This paper presents a state-of-the-art iris segmentation framework specifically for non-ideal irises. The framework adopts coarse-to-fine strategy to localize different boundaries. In the approach, pupil is coarsely detected using an iterative search method exploiting dynamic thresholding and multiple local cues. The limbic boundary is first approximated in polar space using adaptive filters and then refined in Cartesian space. The framework is quite robust and unlike the previously reported works, does not require tuning of parameters for different databases. The segmentation accuracy (SA) is evaluated using well known measures; precision, recall and F-measure, using the publicly available ground truth data for challenging iris databases; CASIAV4-Interval, ND-IRIS-0405, and IITD. In addition, the approach is also evaluated on highly challenging periocular images of FOCS database. The validity of proposed framework is also ascertained by providing comprehensive comparisons with classical approaches as well as state-of-the-art methods such as; CAHT, WAHET, IFFP, GST and Osiris v4.1. The results demonstrate that our approach provides significant improvements in segmentation accuracy as well as in recognition performance that too with lower computational complexity.

Periocular Feature Extraction Based on LBP and DLDA

Periocular recognition is an emerging field of research and people have experimented with some feature extraction techniques to extract robust and unique features from the periocular region. In this paper, we propose a novel feature extraction approach to use periocular region as a biometric trait. In this approach we first applied Local Binary Patterns (LBPs) to extract the texture information from the periocular region of the image and then applied Direct Linear Discriminant Analysis (DLDA) to produce discriminative low-dimensional feature vectors. The approach is evaluated on the UBIRIS v2 database and we achieved 94% accuracy which is a significant improvement in the performance of periocular recognition.

Periocular recognition based on Gabor and Parzen PNN

Recently periocular biometrics has drawn lot of attention of researchers and some efforts have been presented in the literature. In this paper, we propose a novel and robust approach for periocular recognition. In the approach face is detected in still face images which is then aligned and normalized. We utilized entire strip containing both the eyes as periocular region. For feature extraction, we computed the magnitude responses of the image filtered with a filter bank of complex Gabor filters. Feature dimensions are reduced by applying Direct Linear Discriminant Analysis (DLDA). The reduced feature vector is classified using Parzen Probabilistic Neural Network (PPNN). The experimental results demonstrate a promising verification and identification accuracy, also the robustness of the proposed approach is ascertained by providing comprehensive comparison with some of the well known state-of-the-art methods using publicly available face databases; MBGC v2.0, GTDB, IITK and PUT.

Robust periocular biometrics based on Local Phase Quantisation and Gabor transform

In this paper, we propose a novel and robust approach for periocular recognition. Specifically, we propose fusion of Local Phase Quantization(LPQ) and Gabor wavelet descriptors to improve recognition performance and achieve robustness. We have utilized publicly available challenging still face images databases; MBGC v2.0, GTDB, PUT and Caltech. In the approach face is detected and normalized using eye centres. The region around left and right eyes, including eyebrow is extracted as left periocular and right periocular. The LPQ descriptor is then applied to extract the phase statistics features computed locally in a rectangular window. The descriptor is invariant to blur and also to uniform illumination changes. We also computed the Gabor magnitude response of the image, which encodes shape information over a broader range of scales. To reduce dimensionality of the operators and to extract discriminative features, we further utilized DLDA (Direct Linear Discriminant Analysis). The experimental analysis demonstrate that combination of LPQ and Gabor scores provides significant improvement in the performance and robustness, than applied individually.

Collarette Region Recognition based on Wavelets and Direct Linear Discriminant Analysis

recognition is seen as a highly reliable biometric technology. The performance of iris recognition is severely impacted when encountering poor quality images. The selection of the features subset and the classification is an important issue for iris biometrics. Here, we explored the contribution of collarette region in identifying a person. We applied five level haar wavelet decomposition for collarette region feature extraction and used the second level approximation coefficients combined with fifth level vertical coefficients for better accuracy. Then, we applied Direct Linear Discriminant Analysis (DLDA) to produce discriminative low-dimensional feature vectors. The approach is evaluated on CASIA Iris Interval database and we achieved 98.96% accuracy using the collarette region which is a significant improvement in the performance of recognition.

Local Gabor Rank Pattern (LGRP): A novel descriptor for face representation and recognition

The Gabor filters are considered one of the best image representation approaches for face recognition (FR). Researchers have exploited various configurations of Gabor magnitude as well as Gabor phase responses and their modeling with other descriptors. In this paper, we propose a novel face representation approach; Local Gabor Rank Pattern (LGRP), which exploits ordinal ranking of Gabor response images. To take advantage of both magnitude and phase parts, we derive, local Gabor Magnitude Rank Pattern (LGMRP) and local Gabor Phase Rank Pattern (LGPRP) descriptors. We also assign different weights for different LGRPs using variance measure of Gabor coefficients. The descriptors are formed using regional histograms extracted from encoded Gabor filter responses. Furthermore, the LGMRP and LGPRP descriptors are combined within a score-level fusion framework to further improve classification accuracy by maximizing the complementary effect. Extensive experiments on standard face benchmark FERET show that the proposed methods outperform conventional Gabor counterparts as well as other Gabor encoding methods in both constrained and unconstrained FR. The proposed methods also achieve comparable performance to state-of-the-art descriptor based methods in FR.

Color local phase quantization (CLPQ)- A new face representation approach using color texture cues

In this paper, we introduce new methods to encode color local texture features for enhanced face representation. In particular, we first propose a novel descriptor; color local phase quantization (CLPQ), which incorporates (channel-wise) unichrome and (cross channel) opponent features in frequency domain. Furthermore, we extend the CLPQ descriptor to multiple scales i.e. multiscale color LPQ (MS-CLPQ), which exploits the complementary information in different scales. In addition, we extend the multispectral LBP to multiple scales and propose multiscale color LBP (MS-CLBP), which provides illumination invariance and extracts features in spatial domain. To formulate the proposed color local texture descriptors, the unichrome and opponent features are combined using image-level fusion strategy and final representation of the descriptors is obtained using concatenation of regional histograms. To reduce high dimensionality of features, we applied Direct LDA, which also enhances the discrimination ability of the descriptors. The experimental analysis illustrates that proposed MS-CLPQ approach significantly outperforms other descriptor based approaches for face recognition (FR) and score level fusion of MS-CLPQ and MS-CLBP further improves the FR performance and robustness. The validity of the proposed approaches is ascertained by providing comprehensive comparisons on three challenging face databases; FRGC 2.0, GTDB and PUT.

Robust Iris Templates for Efficient Person Identification

Iris recognition is seen as a highly reliable biometric technology. The performance of iris recognition is severely impacted when encountering irises captured in realistic conditions. The selection of the features subset and the classification is an important issue for iris biometrics. In this paper we propose new methods for feature extraction and template creation during enrollment to improve the performance of iris recognition systems. The experiments are based on storing i) multiple templates (template group) for a user ii) Single template by taking average mean of multiple templates iii) Single template calculated from multiple templates using Direct Linear Discriminant Analysis (DLDA). We used CASIA Iris Interval database for our experiments. Experiments report significant improvement in the performance of iris recognition.