Towards a multi-source fusion approach for eye movement-driven recognition

Abstract

26This paper presents a research for the use of multi-source information fusion in the field of eye 27 movement biometrics. In the current state-of-the-art, there are different techniques developed to 28 extract the physical and the behavioral biometric characteristics of the eye movements. In this work, 29we explore the effects from the multi-source fusion of the heterogeneous information extracted by 30 different biometric algorithms under the presence of diverse visual stimuli. We propose a two-stage 31 fusion approach with the employment of stimulus-specific and algorithm-specific weights for fusing 32 the information from different matchers based on their identification efficacy. The experimental 33 evaluation performed on a large database of 320 subjects reveals a considerable improvement in 34 biometric recognition accuracy, with minimal equal error rate (EER) of 5.8%, and best case Rank-1 35 identification rate (Rank-1 IR) of 88.6%. It should be also emphasized that although the concept of 36 multi-stimulus fusion is currently evaluated specifically for the eye movement biometrics, it can be 37 adopted by other biometric modalities too, in cases when an exogenous stimulus affects the extraction 38 of the biometric features. 39 40 Keywords: eye movement biometrics, multi-stimulus fusion, multi-algorithmic fusion 41 42 43 44 45 46 47 48 49 50 51 52The human body provides an invaluable source of distinctive information suitable to be used for the 54 task of biometric recognition [1]. The most well-studied and widely-adopted biometric modalities are 55 the fingerprints, the iris, and the face. Some other explored biometric traits include the palm, the hand 56 geometry, the ears, the nose, and the lips. The analysis of the blood-vessels morphology appears as 57 the main source of biometric features in methods like the vein matching, and the retinal scan. There 58 are also some biometric traits that enfold behavioral characteristics, i.e. traits that are partially 59 connected with the brain activity. Examples of this category involve the speech analysis and voice 60 recognition, the hand-written signature, keystroke dynamics, gait analysis, and the eye movement-61 driven biometrics. Considering the abundance of the existing biometric modalities and the 62 heterogeneity of the associated features, it may come as no surprise that there is a strong trend in the 63 biometric research towards the investigation and adoption of information fusion techniques. 64 Information Fusion in Biometrics 65Information fusion can provide numerous benefits in the domain of biometric recognition. The most 66 obvious among them is the expected performance gain in terms of biometric accuracy due to the 67 combination of evidence gathered from multiple cues [2]. Also, the fusion techniques can be 68 employed for the selection and the promotion of the most informative features among a large set of 69 such features [3]. In addition, the combination of different sources of biometric information can open 70 the path for the creation of b...

Biometric Recognition via Eye Movements

“…These matrices will be combined to generate a single similarity score for each pairwise comparison, which will be used to perform the biometric recognition. In this work, we evaluate four fusion schemes for the combination of similarity scores: the Simple Mean-or Sum-(SM) [Snelick et al 2005], the Weighted Mean based on Rank-1 IR performance (WM) [Rigas et al 2015], the Random Forests (RF) scheme [Breiman 2001], and the Likelihood Ratio (LR) scheme [Nandakumar et al 2008]. To train the parameters of the WM, RF, and LR fusion algorithms, we used the development dataset (see Section 4.1).…”

Reflexive Memory Authenticator: A Proposal for Effortless Renewable Biometrics

“…These rates are generally within 1% and 6-7% for any kind of authentication except for the iris-based ones, which have 0.01% EER but -like most classical biometrics -are not renewable and are vulnerable to theft. Even multimodal biometricswhere one uses multiple sources and biometric features for liveness detection and improved error rates [41] rarely improve below 0.2% EER [8,16,42]. Making the strong assumption that the user data are quite well-distributedwhich is far from guaranteedthis corresponds to a min-entropy below 7 bits, on par with typical password systems.…”

Current research in eye movement biometrics: An analysis based on BioEye 2015 competition

“…Lately, the efficiency of a multi-stimulus and multi-biometric fusion scheme for eye movement-based biometrics was demonstrated by Rigas et al [49]. The performances of different methods for data recorded for diverse visual stimuli ("jumping" point-of-light, text, and video) were efficiently combined via a weighted fusion scheme, for achieving a Rank-1 IR of 88.6% and EER of 5.8% for a large database of 320 subjects.…”

“…Recently, the BioEye 2015 competition [28] was organized with the aim to further advance the conducted research in eye movement biometrics, and to address the emerging challenges in the field. Rigas et al [49], 2015 Multi-source weighted fusion scheme using methods [34], [43], [45] "jumping"-point-of-light, text, video EyeLink 1000 320…”