Tejas I. Dhamecha

Computationally Efficient Face Spoofing Detection with Motion Magnification

For a robust face biometric system, a reliable anti-spoofing approach must be deployed to circumvent the print and replay attacks. Several techniques have been proposed to counter face spoofing, however a robust solution that is computationally efficient is still unavailable. This paper presents a new approach for spoofing detection in face videos using motion magnification. Eulerian motion magnification approach is used to enhance the facial expressions commonly exhibited by subjects in a captured video. Next, two types of feature extraction algorithms are proposed: (i) a configuration of LBP that provides improved performance compared to other computationally expensive texture based approaches and (ii) motion estimation approach using HOOF descriptor. On the Print Attack and Replay Attack spoofing datasets, the proposed framework improves the state-of-art performance, especially HOOF descriptor yielding a near perfect half total error rate of 0%and 1.25% respectively.

On matching latent to latent fingerprints

This research presents a forensics application of matching two latent fingerprints. In crime scene settings, it is often required to match multiple latent fingerprints. Unlike matching latent with inked or live fingerprints, this research problem is very challenging and requires proper analysis and attention. The contribution of this paper is three fold: (i) a comparative analysis of existing algorithms is presented for this application, (ii) fusion and context switching frameworks are presented to improve the identification performance, and (iii) a multi-latent fingerprint database is prepared. The experiments highlight the need for improved feature extraction and processing methods and exhibit large scope of improvement in this important research problem.

Disguise detection and face recognition in visible and thermal spectrums

Face verification, though for humans seems to be an easy task, is a long-standing research area. With challenging covariates such as disguise or face obfuscation, automatically verifying the identity of a person is assumed to be very hard. This paper explores the feasibility of face verification under disguise variations using multi-spectrum (visible and thermal) face images. We propose a framework, termed as Aravrta1, which classifies the local facial regions of both visible and thermal face images into biometric (regions without disguise) and non-biometric (regions with disguise) classes. The biometric patches are then used for facial feature extraction and matching. The performance of the algorithm is evaluated on the IHTD In and Beyond Visible Spectrum Disguise database that is prepared by the authors and contains images pertaining to 75 subjects with different kinds of disguise variations. The experimental results suggest that the proposed framework improves the performance compared to existing algorithms, however there is a need for more research to address this important covariate.

Recognizing disguised faces: human and machine evaluation.

Face verification, though an easy task for humans, is a long-standing open research area. This is largely due to the challenging covariates, such as disguise and aging, which make it very hard to accurately verify the identity of a person. This paper investigates human and machine performance for recognizing/verifying disguised faces. Performance is also evaluated under familiarity and match/mismatch with the ethnicity of observers. The findings of this study are used to develop an automated algorithm to verify the faces presented under disguise variations. We use automatically localized feature descriptors which can identify disguised face patches and account for this information to achieve improved matching accuracy. The performance of the proposed algorithm is evaluated on the IIIT-Delhi Disguise database that contains images pertaining to 75 subjects with different kinds of disguise variations. The experiments suggest that the proposed algorithm can outperform a popular commercial system and evaluates them against humans in matching disguised face images.

On Effectiveness of Histogram of Oriented Gradient Features for Visible to Near Infrared Face Matching

The advent of near infrared imagery and its applications in face recognition has instigated research in cross spectral (visible to near infrared) matching. Existing research has focused on extracting textural features including variants of histogram of oriented gradients. This paper focuses on studying the effectiveness of these features for cross spectral face recognition. On NIR-VIS-2.0 cross spectral face database, three HOG variants are analyzed along with dimensionality reduction approaches and linear discriminant analysis. The results demonstrate that DSIFT with subspace LDA outperforms a commercial matcher and other HOG variants by at least 15%. We also observe that histogram of oriented gradient features are able to encode similar facial features across spectrums.

On incremental semi-supervised discriminant analysis

In various pattern classification problems, semi-supervised discriminant analysis has shown its effectiveness in utilizing unlabeled data to yield better performance than linear discriminant analysis. However, many of these semi-supervised classifiers operate in batch-mode and do not allow to incrementally update the existing model, which is one of the major limitations. This paper presents an incremental semi-supervised discriminant analysis algorithm, which utilizes the unlabeled data for enabling incremental learning. The major contributions of this research are (1) utilizing large unlabeled training set to estimate the total scatter matrix, (2) incremental learning approach that requires updating only the between-class scatter matrix and not the total scatter matrix, and (3) utilizing manifold regularization for robust estimation of total variability and sufficient spanning set representation for incremental learning. Using face recognition as the case study, evaluation is performed on the CMU-PIE, CMU-MultiPIE, and NIR-VIS-2.0 datasets. The experimental results show that the incremental model is consistent with the batch counterpart and reduces the training time significantly. HighlightsIncremental semi-supervised discriminant analysis algorithm is proposed.Large unlabeled data is utilized to estimate total scatter in discriminant analysis.It does not require to incrementally update the total scatter eigenmodel.A face recognition case study is shown on CMU-PIE, NIR-VIS-2.0, and MultiPIE databases.The proposed ISSDA requires significantly less computational time and maintains accuracy.

Incremental subclass discriminant analysis: A case study in face recognition

Subclass discriminant analysis is found to be applicable under various scenarios. However, it is computationally expensive to update the between-class and within-class scatter matrices in batch mode. This research presents an incremental subclass discriminant analysis algorithm to update SDA in incremental manner with increasing number of samples per class. The effectiveness of the proposed algorithm is demonstrated using face recognition in terms of identification accuracy and training time. Experiments are performed on the AR face database and compared with other subspace based incremental and batch learning algorithms. The results illustrate that, compared to SDA, incremental SDA yields significant reduction in time along with comparable accuracy.

Is gender classification across ethnicity feasible using discriminant functions

Over the years, automatic gender recognition has been used in many applications. However, limited research has been done on analyzing gender recognition across ethnicity scenario. This research aims at studying the performance of discriminant functions including Principal Component Analysis, Linear Discriminant Analysis and Subclass Discriminant Analysis with the availability of limited training database and unseen ethnicity variations. The experiments are performed on a heterogeneous database of 8112 images that includes variations in illumination, expression, minor pose and ethnicity. Contrary to existing literature, the results show that PCA provides comparable but slightly better performance compared to PCA+LDA, PCA+SDA and PCA+SVM. The results also suggest that linear discriminant functions provide good generalization capability even with limited number of training samples, principal components and with cross-ethnicity variations.

Annotated crowd video face database

Research in face recognition under constrained environment has achieved an acceptable level of performance. However, there is a significant scope for improving face recognition capabilities in unconstrained environment including surveillance videos. Such videos are likely to record multiple people within the field of view. Face recognition in such a setting poses a set of challenges including unreliable face detection, multiple subjects performing different actions, low resolution, and sensor interoperability. In general, existing video face databases contain one subject in a video sequence. However, real world video sequences are more challenging and generally contain more than one person in a video. Therefore, in this paper, we provide an annotated crowd video face (ACVF-2014) database, along with face landmark information to encourage research in this important problem. The ACVF-2014 dataset contains 201 videos of 133 subjects where each video contains multiple subjects. We provide two distinct use-case scenarios, define their experimental protocols, and report baseline verification results using OpenBR and FaceVACS. The results show that both the baseline results do not yield more than 0.16 genuine accept rate @ 0.01 false accept rate. A software package is also developed to help researchers evaluate their systems using the defined protocols.

Discriminative FaceTopics for face recognition via latent Dirichlet allocation

Latent Dirichlet Allocation is a widely used approach for topic modeling and it has been successfully applied in several information retrieval applications. In this paper, we introduce this modeling technique for face recognition, by making an analogy between the two domains. We utilize latent Dirichlet allocation to represent facial regions in terms of FaceTopics. Further, linear discriminant analysis is utilized to obtain discriminative FaceTopics which are more suitable for classification tasks. The performance of the proposed approach is evaluated on the CMU-MultiPIE dataset under illumination and expression variations. The evaluation on over more than 50k images shows the effectiveness of the proposed approach. Further, the proposed approach shows improved identification results on e-PRIP dataset for matching composite sketches to photos.