Krzysztof Kryszczuk

Face Authentication Competition on the BANCA Database

This paper details the results of a face verification competition [2] held in conjunction with the First International Conference on Biometric Authentication. The contest was held on the publically available BANCA database [1] according to a defined protocol [6]. Six different verification algorithms from 4 academic and commercial institutions submitted results. Also, a standard set of face recognition software from the internet [3] was used to provide a baseline performance measure.

Improving classification with class-independent quality measures: Q-stack in face verification

Existing approaches to classification with signal quality measures make a clear distinction between the single- and multiple classifier scenarios. This paper presents an uniform approach to dichotomization based on the concept of stacking, Q-stack, which makes use of classindependent signal quality measures and baseline classifier scores in order to improve classification in uni- and multimodal systems alike. In this paper we demonstrate the application of Q-stack on the task of biometric identity verification using face images and associated quality measures. We show that the use of the proposed technique allows for reducing the error rates below those of baseline classifiers in single- and multiclassifier scenarios. We discuss how Q-stack can serve as a generalized framework in any single, multiple, and multimodal classifier ensemble.

Quality dependent fusion of intramodal and multimodal biometric experts

We address the problem of score level fusion of intramodal and multimodal experts in the context of biometric identity verification. We investigate the merits of confidence based weighting of component experts. In contrast to the conventional approach where confidence values are derived from scores, we use instead raw measures of biometric data quality to control the influence of each expert on the final fused score. We show that quality based fusion gives better performance than quality free fusion. The use of quality weighted scores as features in the definition of the fusion functions leads to further improvements. We demonstrate that the achievable performance gain is also affected by the choice of fusion architecture. The evaluation of the proposed methodology involves 6 face and one speech verification experts. It is carried out on the XM2VTS data base.

Reliability-based decision fusion in multimodal biometric verification systems

We present a methodology of reliability estimation in the multimodal biometric verification scenario. Reliability estimation has shown to be an efficient and accurate way of predicting and correcting erroneous classification decisions in both unimodal (speech, face, online signature) and multimodal (speech and face) systems. While the initial research results indicate the high potential of the proposed methodology, the performance of the reliability estimation in a multimodal setting has not been sufficiently studied or evaluated. In this paper, we demonstrate the advantages of using the unimodal reliability information in order to perform an efficient biometric fusion of two modalities. We further show the presented method to be superior to state-of-the-art multimodal decision-level fusion schemes. The experimental evaluation presented in this paper is based on the popular benchmarking bimodal BANCA database.

Study of the Distinctiveness of Level 2 and Level 3 Features in Fragmentary Fingerprint Comparison

In this paper we present the results of an experiment which aims to provide an insight into the problems related to the fingerprint recognition from its fragment. Level 2 and Level 3 features are considered, and their distinctive potential is estimated in respect to the considered area of a fingerprint fragment. We conclude that the use of level 3 features can offer at least a comparable recognition potential from a small area fingerprint fragment, as the level 2 features offer for fragments of larger area.

Estimation of the number of clusters using multiple clustering validity indices

One of the challenges in unsupervised machine learning is finding the number of clusters in a dataset. Clustering Validity Indices (CVI) are popular tools used to address this problem. A large number of CVIs have been proposed, and reports that compare different CVIs suggest that no single CVI can always outperform others. Following suggestions found in prior art, in this paper we formalize the concept of using multiple CVIs for cluster number estimation in the framework of multi-classifier fusion. Using a large number of datasets, we show that decision-level fusion of multiple CVIs can lead to significant gains in accuracy in estimating the number of clusters, in particular for high-dimensional datasets with large number of clusters.

Impact of combining quality measures on biometric sample matching

Biometric matching involves a comparison of two biometric data samples. In practical applications, one or both of the samples may be of degraded quality, in respect to the nominal quality of similar samples acquired in controlled conditions. It has been shown in prior art that in such situations, the integration of quality information into the process of bio-metric matching can lead to significantly improved classification performance of the biometric matcher. To facilitate such an integration, quality measures originating from both compared biometric samples are usually combined into one quality score. In this paper, we analyze the merit of doing so. We revisit the problem from a pattern classification perspective, and show that using individual quality measures as separate classification features frequently leads to a superior performance of a biometric system in comparison with the system in which quality measures are mapped into one quality score. We provide experimental support of this claim using synthetic data, as well as real biometric database, on the examples of face, fingerprint and multi-modal matching.

Q-stack: uni- and multimodal classifier stacking with quality measures

The use of quality measures in pattern classification has recently received a lot of attention in the areas where the deterioration of signal quality is one of the primary causes of classification errors. An example of such domain is biometric authentication. In this paper we provide a novel theoretical paradigm of using quality measures to improve both uni- and multimodal classification. We introduce Q - stack, a classifier stacking method in which feature similarity scores obtained from the first classification step are used in ensemble with the quality measures as features for the second classifier. Using two-class, synthetically generated data, we demonstrate how Q - stack helps significantly improve both uni- and multimodal classification in the presence of signal quality degradation.

Detection of Slow Light Level Reduction

Although the increments and decrements in luminance are a fairly well researched and known area in vision science, most of the work has been focused on fast changing stimuli (flicker) or static threshold measurements. Notmuch is known about visual processing of slow-changing stimuli. The described work focuses on measurements of detection threshold for slow reduction of luminance, below the frequencies at which the fluctuations may be referred to as flicker. The experimental study shows that for the used range of luminance reduction rates the detection threshold yields a constant value. Additionally, this value is not influenced by the presence of mental distraction during the detection process.

Credence estimation and error prediction in biometric identity verification

This paper focuses on the estimation of credence in the correctness of classification decisions produced by a biometric identity verification system. We adopt the concept of decision credence defined in terms of subjective Bayesian degree of belief. We demonstrate how credence estimates can be used to predict verification errors and to rectify them, thus improving the classification performance. We also show how the framework of credence estimation helps handle erroneous classification decisions thanks to seamless incorporation of quality measures. Further, we demonstrate that credence information can be effectively applied to perform fusion of decisions in a multimodal scenario.