Paritosh Mittal

Group sparse representation based classification for multi-feature multimodal biometrics

Group sparse representation based classification algorithm is proposed for feature-level multimodal biometrics.The algorithm is able to handle missing features in multimodal scenario.Experimental results on WVU and real world LEA databases show efficacy of the proposed algorithm. Multimodal biometrics technology consolidates information obtained from multiple sources at sensor level, feature level, match score level, and decision level. It is used to increase robustness and provide broader population coverage for inclusion. Due to the inherent challenges involved with feature-level fusion, combining multiple evidences is attempted at score, rank, or decision level where only a minimal amount of information is preserved. In this paper, we propose the Group Sparse Representation based Classifier (GSRC) which removes the requirement for a separate feature-level fusion mechanism and integrates multi-feature representation seamlessly into classification. The performance of the proposed algorithm is evaluated on two multimodal biometric datasets. Experimental results indicate that the proposed classifier succeeds in efficiently utilizing a multi-feature representation of input data to perform accurate biometric recognition.

Composite sketch recognition via deep network - a transfer learning approach

Sketch recognition is one of the integral components used by law enforcement agencies in solving crime. In recent past, software generated composite sketches are being preferred as they are more consistent and faster to construct than hand drawn sketches. Matching these composite sketches to face photographs is a complex task because the composite sketches are drawn based on the witness description and lack minute details which are present in photographs. This paper presents a novel algorithm for matching composite sketches with photographs using transfer learning with deep learning representation. In the proposed algorithm, first the deep learning architecture based facial representation is learned using large face database of photos and then the representation is updated using small problem-specific training database. Experiments are performed on the extended PRIP database and it is observed that the proposed algorithm outperforms recently proposed approach and a commercial face recognition system.

Recognizing composite sketches with digital face images via SSD dictionary

Sketch recognition has important law enforcement applications in detecting and apprehending suspects. Compared to hand drawn sketches, software generated composite sketches are faster to create and require lesser skill sets as well as bring consistency in sketch generation. While sketch generation is one side of the problem, recognizing composite sketches with digital images is another side. This paper presents an algorithm to address the second problem, i.e. matching composite sketches with digital images. The proposed algorithm utilizes a SSD based dictionary generated via 50,000 images from the CMU Multi-PIE database. The gallery-probe feature vectors created using SSD dictionary are matched using GentleBoostKO classifier. The results on extended PRIP composite sketch database show the effectiveness of the proposed algorithm.

Boosting local descriptors for matching composite and digital face images

Sketch recognition is one of the most challenging applications of face recognition. Due to the incorrectness of features in the witness description, standard face recognition algorithms are generally not applicable to matching sketches with digital face images. This research designs a patch based face recognition algorithm that generates patches around fiducial features and extracts local information from these patches using Daisy descriptor. The information extracted from these patches are then efficiently matched using GentleBoostKO algorithm. The experiments performed on the PRIP composite face image database show that the proposed algorithm yields promising results and outperforms existing state-of-the-art algorithms and a commercial system.

Composite sketch recognition using saliency and attribute feedback

Propose composite sketch to photo matching algorithm using visual saliency and combination of texture features.Attributes such as gender and ethnicity information is used to improve recognition performance.Multiple experts such as sketches from different artists are combined for further increasing accuracy. Recent interest and requirement of law enforcement agencies in matching composite sketches with digital images has instigated research in this important face recognition problem. In this paper, we propose feature extraction and matching algorithm using visual saliency and combination of texture features for matching composite sketches with digital photos. The attributes such as gender, ethnicity, and skin color are utilized for re-ordering the ranked list. Further, information from multiple experts such as multiple composite sketch generation tools or artists is combined for improving the matching performance. The results obtained on the extended PRIP database show that the proposed algorithm improves the state-of-art in matching composite sketch and digital face images and yields the rank 50 identification accuracy of 70.3% on a database of 1500 subjects.

Person identification at a distance via ocular biometrics

The performance of iris recognition reduces when the images are captured at a distance. However, such images generally contain periocular region which can be utilized for person recognition. In this research, we propose a novel context switching algorithm that dynamically selects the best descriptor for color iris and periocular regions. Using predefined protocols, the performance of the proposed algorithm is evaluated on UBIRIS V2 and FRGC datasets, and the results show improved performance compared to existing algorithms.

Metadata based recommender systems

For building a recommendation system the eCommerce portal gathers the users ratings on various items in order to determine his/her choice regarding its merchandise. The portal also collects metadata for the user when he/she signs up and becomes a part of the system; therefore the portal has access to information such as users age, gender, occupation, location, etc. Till date almost all prior studies used the metadata for alleviating the cold-start problem; this information was not used for improving the recommendations. For the first time in this work, we propose a simple neighborhood selection technique by giving importance to the metadata groups for improving the recommendations.

At-a-distance person recognition via combining ocular features

Person recognition is a challenging research problem particularly if the images are captured at a distance and only ocular region is present. In this research, we present a framework that extracts multiple features from iris and periocular regions from near infrared images captured at a distance of 2 meters or more. Using these features and random decision forest, fusion and classification is performed and verification results are reported. On CASIA V4-at-a-distance and FOCS databases, the proposed algorithm yields state-of-the-art results; particularly achieving over 61% genuine accept rate at 0.1% false accept rate on complete CASIA V4-at-a-distance database.

A Blind Compressed Sensing formulation for collaborative filtering

For the first we frame collaborative filtering as a Blind Compressed Sensing (BCS) problem. Our formulation stems from the standard matrix factorization approach of decomposing the ratings matrix into a user latent factor matrix and an item latent factor matrix. Previous studies assumed both these matrices to be dense, i.e. both users and items had non-zero values for all factors. This assumption is true for the user matrix but not for the item matrix. It is not possible for an item matrix to have non zero values for all the factors, i.e. the item matrix is likely to be sparse. This assumption leads to the BCS formulation. We compared our proposed method with state-of-the-art matrix factorization approaches; we find that our method yields better accuracy compared to these.