Sanaul Hoque

Evaluating Biometric Encryption Key Generation Using Handwritten Signatures

In traditional cryptosystems, user authentication is based on the possession of secret keys/tokens. Such keys can be forgotten, lost, stolen, or may be illegally shared, but an ability to relate a cryptographic key to biometric data can enhance the trustworthiness of a system. In this paper, we demonstrate how biometric keys can be generated directly from live biometrics, under certain conditions, by partitioning feature space into subspaces and partitioning these into cells, where each cell subspace contributes to the overall key generated. We evaluate the proposed scheme on real biometric data, representing both genuine samples and attempted imitations. Experimental results then demonstrate the extent to which the proposed technique can be implemented reliably in possible practical scenarios.

A new chain-code quantization approach enabling high performance handwriting recognition based on multi-classi .er schemes

In this paper initially we propose a novel approach toclassify handwritten characters based on a directional decompositionof the corresponding chain-code representation.This is alternative to previous transformations of thechain-codes proposed by the authors, namely the orderedand random decomposition of the bit-planes resulting fromthe binary representation of the chain-codes. Subsequentlywe utilize the power of the recently developed multiple classifierschemes using sntuple classifiers to integrate the complimentaryinformation encapsulated in all three transformationsinto a more powerful and robust character recognitionsystem. The results obtained through a series ofcross-validation experiments show that the proposed fusionscheme not only outperforms its constituent parts and anumber of other successful classifiers, but also enables significantsavings in memory requirements compared to theoriginal sntuple-based recognition system.

Genetic Algorithms for Multi-classifier System Configuration: A Case Study in Character Recognition

We describe a multiple classifier system which incorporates an automatic self-configuration scheme based on genetic algorithms. Our main interest in this paper is focused on exploring the statistical properties of the resulting multi-expert configurations. To this end we initially test the proposed system on a series of tasks of increasing difficulty drawn from the domain of character recognition. We then proceed to investigate the performance of our system not only in comparison to that of its constituent classifiers, but also in comparison to an independent set of individually optimised classifiers. Our results illustrate that significant gains can be obtained by integrating a genetic algorithm based optimisation process into multi-classifier schemes both in the performance enhancement and in the reduction of its volatility, especially as the task domain becomes more complex.

Trainable Multiple Classifier Schemes for Handwritten Character Recognition

In this paper we propose two novel multiple classifier fusion schemes which, although different in terms of architecture, share the idea of dynamically extracting additional statistical information about the individually trained participant classifiers by reinterpreting their outputs on a validation set. This is achieved through training on the resulting intermediate feature spaces of another classifier, be it a combiner or an intermediate stage classification device. We subsequently implemented our proposals as multi-classifier systems for handwritten character recognition and compare the performance obtained through a series of cross-validation experiments of increasing difficulty. Our findings strongly suggest that both schemes can successfully overcome the limitations imposed on fixed combination strategies from the requirement of comparable performance levels among their participant classifiers. In addition, the results presented demonstrate the significant gains achieved by our proposals in comparison withb oth individual classifiers experimentally optimized for the task in hand, and a multi-classifier system design process which incorporates artificial intelligence techniques.

Are Two Eyes Better than One? An Experimental Investigation on Dual Iris Recognition

Iris recognition using both eyes of an individual has not been extensively researched in the available literature. However, for iris recognition at a distance, capturing a good quality image of the same eye every time is a challenging task and a dual iris approach is potentially beneficial. In the present work, we investigate the advantages of using dual iris approach for iris recognition, comparing the performances of using both eyes with those obtained using only one eye.

Bit plane decomposition and the scanning n-tuple classifier

This paper describes a multiple classifier configuration for high performance off-line handwritten character recognition applications. Along with a conventional scanning n-tuple classifier (or sn-tuple) implementation, three other sn-tuple systems have been used which are trained using a binary feature set extracted from the contour chain-codes using a novel decomposition technique. The overall accuracy thus achievable by the proposed scheme is much higher than most other classification systems available and the added complexity (over conventional sn-tuple system) is minimal.

A review of information fusion techniques employed in iris recognition systems

Iris recognition has shown to be one of the most reliable biometric authentication methods. The majority of iris recognition systems which have been developed require a constrained environment to enrol and recognise the user. If the user is not cooperative or the capture environment changes then the accuracy of the iris recognition system may decrease significantly. To minimise the effect of such limitations, possible solutions include the use of multiple channels of information such as using both eyes or extracting more iris feature types and subsequently employing an efficient fusion method. In this paper, we present a review of iris recognition systems using information from multiple sources that are fused in different ways or at different levels. A categorisation of the iris recognition systems incorporating multiple classifier systems is also presented. As a new desirable dimension of a biometric system, besides those proposed in the literature, the mobility of such a system is introduced in this work. The review charts the path towards greater flexibility and robustness of iris recognition systems through the use of information fusion techniques and points towards further developments in the future leading to mobile and ubiquitous deployment of such systems.

A Survey of Point-Source Specular Reflections in Noisy Iris Images

This paper presents an examination of a selection of images taken from the UBIRIS.v2 dataset to explore the characteristics of point-source reflections present in the images. These reflections were some of the most commonly found sources of noise in iris images acquired under visual wavelength light and clearly impact the accuracy of iris recognition systems. The spatial and intensity distributions of these reflections is studied and results are presented that can be used to model their behaviour. This information can be helpful for developing more accurate iris synthesis techniques and for the study of iris image focus assessment as well as developing better matching algorithms for iris recognition.

The autonomous document object (ADO) model

A novel strategy for the representation and manipulation of multimedia documents is introduced. The model allows such documents to be treated as autonomous entities capable of representing and modifying their own internal structures. The model is sufficiently flexible as to allow any given arbitrary document format and any programming language to be incorporated within its implementation.

Task sensitivity in EEG biometric recognition

This work explores the sensitivity of electroencephalographic-based biometric recognition to the type of tasks required by subjects to perform while their brain activity is being recorded. A novel wavelet-based feature is used to extract identity information from a database of 109 subjects who performed four different motor movement/imagery tasks while their data were recorded. Training and test of the system was performed using a number of experimental protocols to establish if training with one type of task and tested with another would significantly affect the recognition performance. Also, experiments were conducted to evaluate the performance when a mixture of data from different tasks was used for training. The results suggest that performance is not significantly affected when there is a mismatch between training and test tasks. Furthermore, as the amount of data used for training is increased using a combination of data from several tasks, the performance can be improved. These results indicate that a more flexible approach may be incorporated in data collection for EEG-based biometric systems which could facilitate their deployment and improved performance.