Daigo Muramatsu

This paper describes a method of discriminant analysis for cross-view recognition with a relatively small number of training samples. Since appearance of a recognition target (e.g., face, gait, gesture, and action) is in general drastically changes as an observation view changes, we introduce multiple view-specific projection matrices and consider to project a recognition target from a certain view by a corresponding view-specific projection matrix into a common discriminant subspace. Moreover, conventional vectorized representation of an originally higher-order tensor object (e.g., a spatio-temporal image in gait recognition) often suffers from the curse of dimensionality dilemma, we therefore encapsulate the multiple view-specific projection matrices in a framework of discriminant analysis with tensor representation, which enables us to overcome the curse of dimensionality dilemma. Experiments of cross-view gait recognition with two publicly available gait databases show the effectiveness of the proposed method in case where a training sample size is small.

Gait recognition is a useful biometric trait for person authentication because it is usable even with low image resolution. One challenge is robustness to a view change (cross-view matching); view transformation models (VTMs) have been proposed to solve this. The VTMs work well if the target views are the same as their discrete training views. However, the gait traits are observed from an arbitrary view in a real situation. Thus, the target views may not coincide with discrete training views, resulting in recognition accuracy degradation. We propose an arbitrary VTM (AVTM) that accurately matches a pair of gait traits from an arbitrary view. To realize an AVTM, we first construct 3D gait volume sequences of training subjects, disjoint from the test subjects in the target scene. We then generate 2D gait silhouette sequences of the training subjects by projecting the 3D gait volume sequences onto the same views as the target views, and train the AVTM with gait features extracted from the 2D sequences. In addition, we extend our AVTM by incorporating a part-dependent view selection scheme (AVTM_PdVS), which divides the gait feature into several parts, and sets part-dependent destination views for transformation. Because appropriate destination views may differ for different body parts, the part-dependent destination view selection can suppress transformation errors, leading to increased recognition accuracy. Experiments using data sets collected in different settings show that the AVTM improves the accuracy of cross-view matching and that the AVTM_PdVS further improves the accuracy in many cases, in particular, verification scenarios.

This paper describes a method for score-level fusion in multi-cue two-class classification problems. Fusion based on the probability density function (PDF) of multiple scores given for each class is a promising approach because it guarantees optimality as long as the estimated PDFs are correct. Instead of lattice-type control points used in previous non-parametric density-based approaches, floating control points (FCPs) are introduced to improve scalability and the whole posterior distribution is represented by interpolation or extrapolation using generalized Delaunay triangulation. Given a set of FCPs obtained by k-means, posteriors on the FCPs are estimated by an energy minimization framework using training samples. The experiments, using both simulation data as well as several types of real data from three publicly available score databases for multi-cue biometric authentication, demonstrate the effectiveness of the proposed method.

Gait is a unique and promising behavioral biometrics which allows to authenticate a person even at a distance from the camera. Since a matching pair of gait features are often drawn from different views due to differences in camera position/attitude and walking directions in the real world, it is important to cope with cross-view gait recognition. In this paper, we propose a discriminative approach to cross-view gait recognition using view-dependent projection matrices, unlike the existing discriminant approaches which utilize only a single common projection matrix for different views. We demonstrated the effectiveness of the proposed method through cross-view gait recognition experiments with two publicly available gait datasets. In addition, since the success of the discriminant analysis relies on the training sample size, we show the effect of transfer learning across two gait datasets as well as provide the rigorous sensitivity analysis of the proposed method against the number of training subjects ranging from 10 to approximately 1,000 subjects.

Gait features can be extracted from low-quality image sequences captured at a distance, which makes gait recognition a useful method in forensics. However, the accuracy of gait recognition is often degraded in a cross-view setting, which often occurs in forensic cases. Therefore, we propose a gait recognition algorithm that achieves high accuracy in a cross-view setting. In this paper, we focus on a view transformation model-based approach, extract transformation consistency measures, and propose to use these measures for cross-view recognition. To evaluate the accuracy of the proposed method, we draw receiver operation characteristic curves together with Tippett plots, and evaluate discrimination ability and calibration quality. The experimental results show that our proposed method achieves good results in terms of discrimination and calibration.

This paper describes a quality-dependent score-level fusion framework of face, gait, and the height biometrics from a single walking image sequence. Individual person authentication accuracies by face, gait, and the height biometrics, are in general degraded when spatial resolution (image size) and temporal resolution (frame-rate) of the input image sequence decrease and the degree of such accuracy degradation differs among the individual modalities. We therefore set the optimal weights of the individual modalities based on linear logistic regression framework depending on a pair of the spatial and temporal resolutions, which are called qualities in this paper. On the other hand, it is not a realistic solution to compute and store the optimal weights for all the possible qualities in advance, and also the optimal weights change across the qualities in a nonlinear way. We thus propose a method to estimate the optimal weights for arbitrary qualities from a limited training pairs of the optimal weights and the qualities, based on Gaussian process regression with a nonlinear kernel function. Experiments using a publicly available large population gait database with 1,935 subjects under various qualities, showed that the person authentication accuracy improved by successfully estimating the weights depending on the qualities.

We focus on gait recognition for criminal investigation. In criminal investigation, person authentication is performed by comparing target data at the crime scene and multiple gait data with slightly different views from that of the target data. For this task, we propose fusion of direct cross-view matching. Cross-view matching generally produces worse result than those of same-view matching when view-variant features are used. However, the correlation between cross-view matching with different view pairs is low and it provides improved accuracy. Experimental results performed utilizing large-scale dataset under settings resembling actual criminal investigation cases, show that the proposed approach works well. © 2013 Information Processing Society of Japan.

View difference is a factor that degrades the accuracy of gait recognition. A solution to reducing accuracy degradation is applying a view transformation model (VTM) that encodes a joint subspace of multi-view gait features trained from multiple training subjects. In the VTM framework, once an intrinsic vector of a test subject in the joint subspace is estimated from a gait feature with a source view (e.g. probe view), a gait feature with a destination view (e.g. gallery view) is generated for the same-view matching. Although this family of methods can improve the total accuracy, the quality of generated gait features depends on a test gait feature, and may be relevant to the accuracy of gait recognition. We therefore propose a method of incorporating the quality measure of the generated gait feature into the VTM framework. We employ the projection error into the joint subspace as the quality measure. A posterior probability is then computed by incorporating the quality measure. The accuracy evaluation against a subset of a public database collected from 1,912 subjects shows that the proposed method further improves the accuracy. © 2013 IEEE.

This paper describes a method for multi-view multimodal biometrics from a single walking image sequence. As multi-modal cues, we adopt not only face and gait but also the actual height of a person, all of which are simultaneously captured by a single camera. As multi-view cues, we use the variation in the observation views included in a single image sequence captured by a camera with a relatively wide field of view. This enables us to improve the authentication of a person based on multiple modalities and views, while retaining the potential for real applications such as surveillance and forensics using only a single image sequence (a single session with a single camera). In the experiments, we constructed a large-scale multi-view multimodal score data set with 1,912 subjects, and evaluated the proposed method using the data set in a statistically reliable way. We achieved 0.37% equal error rates for the false acceptance and rejection rates in the verification scenarios, and 99.15% rank-1 identification rate in the identification scenarios. © 2013 IEEE.

This paper describes a method of gait recognition using multiple gait features in conjunction with score-level fusion techniques. More specifically, we focus on the state-of-the-art period-based gait features such as a gait energy image, a frequency-domain feature, a gait entropy image, a chronogait image, and a gait flow image. In addition, we employ various types of the score-level fusion approaches including not only conventional transformation-based approaches (e.g., sum-rule and min-rule) but also classification-based approaches (e.g., support vector machine) and density-based approaches (e.g., Gaussian mixture model, kernel density estimation, linear logistic regression). In experiments, the large-population gait database with 3,249 subjects was used to measure the performance improvement in a statistically reliable way. The experimental results show 7% relative improvement on average with regard to equal error rate of the false acceptance rate and false rejection rate in verification scenarios, and also show 20% reduction of the number of candidates to be checked under 1% misdetection rate on average in screening tasks. © 2013 IEEE.

Cameras are used for acquiring online signature data. A camera can acquire not only pen tip trajectories, but also pen holding style. Therefore, we can use information associated with the pen holding style together with the pen tip trajectory for online signature verification. In the work described here, we used a silhouette of a skin region as a feature of pen holding style, and an averaged silhouette was used together with online signatures for verification. Preliminary experiments were performed using data collected from 22 subjects. The experimental results show that the accuracy was improved by combining the online signatures with silhouette-based pen holding style.

We constructed gait verification system for criminal investigation. The system is designed so that criminal investigators can use it and obtain professional gait verification results. We think the system can support criminal investigation where gait can be a clue of the perpetrator's identity. We summarize the constructed system in this paper. © 2013 IEEE.

We investigated the false accept rates against randomly forged signatures written by each signer by using a DTW-based online signature verification algorithm. The experimental results show that attacks using some signers' signatures are stronger than those using skilled forged signatures and that some signers' signatures can be a wolf against the algorithm considered.

The first gait-based person-verification system that can analyze the gait for forensic science is presented. There are many security cameras set in many places, and gait image sequences can be extracted from the images taken by the cameras. The gait can provide information for identity determination in forensic science however, the limitation of gait-based person authentication is that the analysis of gait features is a difficult task for non-specialists of gait person authentication. Our system allows non-specialists to analyze gait images taken in different circumstances, and can provide useful information for criminal investigations. We performed evaluation experiments to evaluate the usability of the constructed system. In the experiments, nine participants used the system to analyze gait features. The experimental results show that novices can use the constructed system correctly and obtain reasonable results. © 2012 IEEE.

Camera-based gait recognition is a useful method for authenticating a person from a distance, even if the resolution of the acquired images is not high. However, different views of the compared gallery and probe decrease the recognition accuracy. To solve this problem, we propose a gait based authentication method that uses an arbitrary view transformation scheme. The proposed method constructs a transformation matrix associated with the view of the set of gallery and probe using a 3D gait database composed of non-target multiple subjects' visual hulls. This matrix is used to transform the gallery gait features with varying views into features with the same view as the probe. Using this scheme, we can minimize the impact of the view difference. We evaluated the accuracy of the proposed method using a gait database composed of 52 subjects. The experimental results show that the proposed method is promising. © 2012 IEEE.

This paper describes a method of score-level fusion to optimize a Receiver Operating Characteristic (ROC) curve for multimodal biometrics. When the Probability Density Functions (PDFs) of the multimodal scores for each client and imposter are obtained from the training samples, it is well known that the isolines of a function of probabilistic densities, such as the likelihood ratio, posterior, or Bayes error gradient, give the optimal ROC curve. The success of the probability density-based methods depends on the PDF estimation for each client and imposter, which still remains a challenging problem. Therefore, we introduce a framework of direct estimation of the Bayes error gradient that bypasses the troublesome PDF estimation for each client and imposter. The lattice-type control points are allocated in a multiple score space, and the Bayes error gradients on the control points are then estimated in a comprehensive manner in the energy minimization framework including not only the data fitness of the training samples but also the boundary conditions and monotonic increase constraints to suppress the over-training. The experimental results for both simulation and real public data show the effectiveness of the proposed method. © 2011 IEEE.

We propose a visual-based online signature verification system. The input module of the system consists of only low-cost cameras (webcams) and does not need an electronic tablet. Online signature data are obtained from the images captured by the webcams by tracking the pen tip. The pen tip tracking is implemented by the sequential Monte Carlo method. Then, the distance between the input signature data and reference signature data enrolled in advance is computed. Finally, the input signature is classified as genuine or a forgery by comparing the distance with a threshold. In this paper, we consider seven camera positions. We performed experiments using a private database consisting of 150 genuine signatures to decide the best camera position. The experimental results show that we should place the webcam to the side of the hand. Moreover, we evaluated the proposed system with a camera placed to the side of the hand against a different database consisting of 390 genuine signatures and 1560 skilled forged signatures. The proposed system achieved an equal error rate of 4.1% against this database. (C) 2009 Elsevier Ltd. All rights reserved.

Attacks using hill-climbing methods have been reported as a vulnerability of biometric authentication systems. In this paper, we propose a robust online signature verification algorithm against such attacks. Specifically, the attack considered in this paper is a hill-climbing forged data attack. Artificial forgeries are generated offline by using the hill-climbing method, and the forgeries are input to a target system to be attacked. In this paper, we analyze the menace of hill-climbing forged data attacks using six types of hill-climbing forged data and propose a robust algorithm by incorporating the hill-climbing method into an online signature verification algorithm. Experiments to evaluate the proposed system were performed using a public online signature database. The proposed algorithm showed improved performance against this kind of attack.

The manner of holding a pen is distinctive among people. Therefore, pen holding style is useful for person authentication. In this paper, we propose a biometric person authentication method using features extracted from images of pen holding style. Images of the pen holding style are captured by a camera, and several features are extracted from the captured images. These features are compared with a reference dataset to calculate dissimilarity scores, and these scores are combined for verification using a three-layer perceptron. Preliminary experiments were performed by using a private database. The proposed system yielded an equal error rate (EER) of 2.6% against an impersonate attack..

The manner of holding a pen is distinctive among people. Therefore, pen holding style is useful for person authentication. In this paper, we propose a biometric person authentication method using features extracted from images of pen holding style. Images of the pen holding style are captured by a camera, and several features are extracted from the captured images. These features are compared with a reference dataset to calculate dissimilarity scores, and these scores are combined for verification using a three-layer perceptron. Preliminary experiments were performed by using a private database. The proposed system yielded an equal error rate (EER) of 2.6%.

A camera-based online signature verification system is proposed in this paper. One web camera is used for data acquisition, and a sequential Monte Carlo method is used for tracking a pen tip. Several distances are computed from an online signature, and a fusion model trained by using AdaBoost combines the distances and computes a final score. Preliminary experiments were performed by using a private database. The proposed system yielded an equal error rate (EER) of 4.0%. © 2009 IEEE.

Several online signature verification systems that rise cameras have been proposed. These systems obtain online signature data from video images by tracking the pen tip. Such systems are very useful because special devices such as pen-operated digital tablets are not; necessary. One drawback, however, is that if the captured images are blurred, pen tip tracking may fail, which causes performance degradation. To solve this problem, here we propose a. scheme to detect such images and re-estimate the pen tip position associated with the blurred images. Our pen tracking algorithm is implemented by rising lie sequential Monte Carlo method, and a sequential marginal likelihood is used for blurred image detection. Preliminary experiments were performed using private data, consisting of 390 genuine signatures and 1560 forged signatures. The experimental results show that the proposed algorithm improved performance in terms of verification accuracy.

Authentication of handwritten signatures is becoming increasingly important. With a rapid increase in the number of people who access Tablet PCs and PDAs, online signature verification is one of the most promising techniques for signature verification. This paper proposes a new algorithm that performs a Monte Carlo based Bayesian scheme for online signature verification. The new algorithm consists of a learning phase and a testing phase. In the learning phase, semi-parametric models are trained using the Markov Chain Monte Carlo (MCMC) technique to draw posterior samples of the parameters involved. In the testing phase, these samples are used to evaluate the probability that a signature is genuine. The proposed algorithm achieved an EER of 1.2% against the MCYT signature corpus where random forgeries are used for learning and skilled forgeries are used for evaluation. An experimental result is also reported with skilled forgery data for learning.

For Pen-input on-line signature verification algorithms, the influence of intersession variability is a considerable problem because hand-written signatures change with time, causing performance degradation. In our previous work, we proposed a user-generic model using AdaBoost. However, this model did not allow for the fact that features of signatures change over time. In this paper, we propose a template renewal method to reduce the performance degradation caused by signature changes over time. In our proposed method, the oldest template is replaced with a new one if the new signature data gives rise to an index which exceeds a threshold value. No further learning is necessary. A preliminary experiment was conducted on a subset of the MCYT database.

The purpose of this project is two fold. The first purpose is to reduce the memory size of our previous handwriting recognition algorithm based on an HMM using Self-Organizing Map (SOM) density tying. The second is to improve recognition capability by incorporating additional information. SOM density tying reduced the dictionary size to 1/7 of the original size, with a recognition rate of 90.45%, only slightly less than the original recognition rate of 91.51%. Our additional feature increased recognition capability to 91.34%.

Authentication of individuals is rapidly becoming an important issue. This paper proposes a new nonlinear algorithm for pen-input on-line signature verification incorporating pen-position, pen-pressure and peninclinations trajectories. A preliminary experiment is performed on a database consisting of 1849 genuine signatures and 3174 skilleddagger forgery signatures from fourteen individuals. Since no fine tuning was done, this preliminary result looks very promising.

Based on a cascode-driver source-follower buffer, and a passive sampling architecture, we have implemented a differential sample-and-hold circuit in a 0.8 μm digital CMOS process. The buffer which eliminates channel length modulation of the driver device behaves very linearly, in low frequencies or sampled-data applications. This is the main reason that this first open-loop CMOS sample-and-hold can achieves very high linearity while functions at very high sampling rate. The circuit achieved -61 dB THD for a 1.42 Vp-p 10 MHz input signal at a 103 MHz sampling rate and -55.9 dB THD for a 1.22 Vp-p 20 MHz at a 101 MHz sampling rate.