Vikas Gottemukkula

Fusing iris and conjunctival vasculature: Ocular biometrics in the visible spectrum

Ocular biometrics refers to the imaging and use of characteristic features of the eyes for personal identification. Traditionally, the iris has been viewed as a powerful ocular biometric cue. However, the iris is typically imaged in the near infrared (NIR) spectrum. RGB images of the iris, acquired in the visible spectrum, offer limited biometric information for dark-colored irides. In this work, we explore the possibility of performing ocular biometric recognition in the visible spectrum by utilizing the iris in conjunction with the vasculature observed in the white of the eye. We design a weighted fusion scheme to combine the information originating from these two modalities. Experiments on a dataset of 50 subjects indicate that such a fusion scheme improves the equal error rate by a margin of 4.5% over an iris-only approach.

A video-based hyper-focal imaging method for iris recognition in the visible spectrum

We design and implement a hyper-focal imaging system for acquiring iris images in the visible spectrum. The proposed system uses a DSLR Canon T2i camera and an Okii controller to capture videos of the ocular region at multiple focal lengths. The ensuing frames are fused in order to yield a single image with higher fidelity. Further, the proposed setup extends the imaging depth-of-field (DOF), thereby preempting the need for employing expensive cameras for increased DOF. Experiments convey the benefits of utilizing a hyper-focal system over a traditional fixed-focus system for performing iris recognition in visible spectrum.

ICIP 2016 competition on mobile ocular biometric recognition

With the unprecedented mobile technology revolution, a number of ocular biometric based personal recognition schemes have been proposed for mobile use cases. The aim of this competition is to evaluate and compare the performance of mobile ocular biometric recognition schemes in visible light on a large scale database (VISOB Dataset ICIP2016 Challenge Version) using standard evaluation methods. Four different teams from universities across the world participated in this competition, submitting five algorithms altogether. The submitted algorithms applied different texture analysis in a learning or a non-learning based framework for ocular recognition. The best results were obtained by a team from Norwegian Biometrics Laboratory (NTNU, Norway), achieving an Equal Error Rate of 0.06% over a quarantined test set.

Classification-guided feature selection for NIRS-based BCI

Motor movements induce distinct patterns in the hemodynamics of the motor cortex, which may be captured by Near-Infrared Spectroscopy (NIRS) for Brain Computer Interfaces (BCI). We present a classification-guided (wrapper) method for time-domain NIRS feature extraction to classify left and right hand movements. Four different wrapper methods, based on univariate and multivariate ranking and sequential forward and backward selection, along with three different classifiers (k-Nearest neighbor, Bayes, and Support Vector Machines) were studied. Using NIRS data from two subjects we show that a rank-based wrapper in conjunction with polynomial SVMs can achieve 100% sensitivity and specificity separating left and right hand movements (5-fold cross validation). Results show the promise of wrapper methods in classifying NIRS signals for BCI applications.

A texture-based method for identificaiton of retinal vasculature

Noting the advantages of texture-based features over the structural descriptors of vascular trees, we investigated texture-based features from gray level cooccurrence matrix (GLCM) and various wavelet packet energies to classify retinal vasculature for biometric identification. Wavelet packet energy features were generated by Daubechies, Coiflets and Reverse Biorthogonal wavelets. Two different entropy methods, Shannon and logarithm of energy, were used to prune wavelet packet decomposition trees. Next, wrapper methods were used for classification-guided feature selection. Features were ranked based on area under the receiver operating curves, Bhattacharya, and t-test metrics. Using the ranked lists, wrapper methods were used in conjunction with Naïve Bayesian, k-nearest neighbor (k-NN), and Support Vector Machine (SVM) classifiers. Best results were achieved by using features from Reverse Biorthogonal 2.4 wavelet packet decomposition in conjunction with a nearest neighbor classifier, yielding a 3-fold cross validation accuracy of 99.42% with a sensitivity and specificity of 98.33% and 99.47% respectively.

Enhanced obfuscation for multi-part biometric templates

Biometrie authentication is being exceedingly utilized into mobile devices as an alternative to passwords. However, for security and privacy reasons, it is important to protect the biometric template. In this paper, we introduce a method to obfuscate and match certain biometric templates comprised of multiple local descriptors derived around spatial interest points. Obfuscation starts by insertion of chaff (fake) interest points along with their respective synthesized descriptors that are statistically similar to original descriptors. Fusion of local matches along with global outlier rejection is used to mitigate the impact of the resulting obfuscation on biometric matching accuracy. The efficacy of the proposed method is demonstrated through ocular and face recognition experiments.