Search for: All records

Many developers of biometric systems start with modest samples before general deployment. However, they are interested in how their systems will work with much larger samples. To assist them, we evaluated the effect of gallery size on biometric performance. Identification rates describe the performance of biometric identification, whereas ROC-based measures describe the performance of biometric authentication (verification). Therefore, we examined how increases in gallery size affected identification rates (i.e., Rank-1 Identification Rate, or Rank-1 IR) and ROC-based measures such as equal error rate (EER). We studied these phenomena with synthetic data as well as real data from a face recognition study. It is well known that the Rank-1 IR declines with increasing gallery size, and that the relationship is linear against log(gallery size). We have confirmed this with synthetic and real data. We have shown that this decline can be counteracted with the inclusion of additional information (features) for larger gallery sizes. We have also described the curves which can be used to predict how much additional information would be required to stabilize the Rank-1 IR as a function of gallery size. These equations are also linear in log(gallery size). We have also shown that the entire ROC-curve was not systematically affected by gallery size, and so ROC-based scalar performance metrics such as EER are also stable across gallery size. Unsurprisingly, as additional uncorrelated features are added to the model, EER decreases. We were interested in determining the impact of adding more features on the median, spread and shape of similarity score distributions. We present evidence that these decreases in EER are driven primarily by decreases in the spread of the impostor similarity score distribution. 

Iris-based biometric authentication is a wide-spread biometric modality due to its accuracy, among other benefits. Improving the resistance of iris biometrics to spoofing attacks is an important research topic. Eye tracking and iris recognition devices have similar hardware that consists of a source of infra-red light and an image sensor. This similarity potentially enables eye tracking algorithms to run on iris-driven biometrics systems. The present work advances the state-of-the-art of detecting iris print attacks, wherein an imposter presents a printout of an authentic user’s iris to a biometrics system. The detection of iris print attacks is accomplished via analysis of the captured eye movement signal with a deep learning model. Results indicate better performance of the selected approach than the previous state-of-the-art. 

Abstract This manuscript presents GazeBase, a large-scale longitudinal dataset containing 12,334 monocular eye-movement recordings captured from 322 college-aged participants. Participants completed a battery of seven tasks in two contiguous sessions during each round of recording, including a – (1) fixation task, (2) horizontal saccade task, (3) random oblique saccade task, (4) reading task, (5/6) free viewing of cinematic video task, and (7) gaze-driven gaming task. Nine rounds of recording were conducted over a 37 month period, with participants in each subsequent round recruited exclusively from prior rounds. All data was collected using an EyeLink 1000 eye tracker at a 1,000 Hz sampling rate, with a calibration and validation protocol performed before each task to ensure data quality. Due to its large number of participants and longitudinal nature, GazeBase is well suited for exploring research hypotheses in eye movement biometrics, along with other applications applying machine learning to eye movement signal analysis. Classification labels produced by the instrument’s real-time parser are provided for a subset of GazeBase, along with pupil area. 

Typically, the position error of an eye-tracking device is measured as the distance of the eye-position from the target position in two-dimensional space (angular offset).  Accuracy is the mean angular offset.  The mean is a highly interpretable measure of central tendency if the underlying error distribution is unimodal and normal. However, in the context of an underlying multimodal distribution, the mean is less interpretable. We will present evidence that the majority of such distributions are multimodal.  Only 14.7% of fixation angular offset distributions  were  unimodal, and  of  these,  only  11.5%  were normally distributed.  (Of the entire dataset, 1.7% were unimodal and normal.)  This multimodality is true even if there is only a single, continuous tracking fixation segment per trial. We present several approaches to measure accuracy in the face of multimodality. We also address the role of fixation drift in partially explaining multimodality. 

It is generally accepted that relatively more permanent (i.e., more temporally persistent) traits are more valuable for biometric performance than less permanent traits. Although this finding is intuitive, there is no current work identifying exactly where in the biometric analysis temporal persistence makes a difference. In this paper, we answer this question. In a recent report, we introduced the intraclass correlation coefficient (ICC) as an index of temporal persistence for such features. Here, we present a novel approach using synthetic features to study which aspects of a biometric identification study are influenced by the temporal persistence of features. What we show is that using more temporally persistent features produces effects on the similarity score distributions that explain why this quality is so key to biometric performance. The results identified with the synthetic data are largely reinforced by an analysis of two datasets, one based on eye-movements and one based on gait. There was one difference between the synthetic and real data, related to the intercorrelation of features in real data. Removing these intercorrelations for real datasets with a decorrelation step produced results which were very similar to that obtained with synthetic features. 

Texture-based features computed on eye movement scan paths have recently been proposed for eye movement biometric applications. Feature vectors were extracted within this prior work by computing the mean and standard deviation of the resulting images obtained through application of a Gabor filter bank. This paper describes preliminary work exploring an alternative technique for extracting features from Gabor filtered scan path images. Namely, features vectors are obtained by downsampling the filtered images, thereby retaining structured spatial information within the feature vector. The proposed technique is validated at various downsampling scales for data collected from 94 subjects during free-viewing of a fantasy movie trailer. The approach is demonstrated to reduce EER versus the previously proposed statistical summary technique by 11.7% for the best evaluated downsampling parameter. 

The majority of eye-tracking systems require user-specific calibration to achieve suitable accuracy. Traditional calibration is performed by presenting targets at fixed locations that form a certain coverage of the device screen. If simple regression methods are used to learn a gaze map from the recorded data, the risk of overfitting is minimal. This is not the case if a gaze map is formed using neural networks, as is often employed in photosensor oculography (PSOG), which raises the question of careful design of calibration procedure. This paper evaluates different calibration data parsing approaches and the collection time-performance trade-off effect of grid density to build a calibration framework for PSOG with the use of video-based simulation framework.