Deep learning techniques are nowadays the leading approaches to solve complex machine learning and pattern recognition problems. For the first time, we utilize state-of-the-art semantic segmentation CNNs to extract vein patterns from near-infrared finger imagery and use them as the actual vein features in biometric finger-vein recognition. In this context, beside investigating the impact of training data volume, we propose a training model based on automatically generated labels, to improve the recognition performance of the resulting vein structures compared to (i) network training using manual labels only, and compared to (ii) well established classical recognition techniques relying on publicly available software. Proposing this model we also take a crucial step in reducing the amount of manually annotated labels required to train networks, whose generation is extremely time consuming and error-prone. As further contribution, we also release human annotated ground-truth vein pixel labels (required for training the networks) for a subset of a well known finger-vein database used in this work, and a corresponding tool for further annotations.
The greatest threat towards securing the organization and its assets are no longer the attackers attacking beyond the network walls of the organization but the insiders present within the organization with malicious intent. Existing approaches helps to monitor, detect and prevent any malicious activities within an organization's network while ignoring the human behavior impact on security. In this paper we have focused on user behavior profiling approach to monitor and analyze user behavior action sequence to detect insider threats. We present an ensemble hybrid machine learning approach using Multi State Long Short Term Memory (MSLSTM) and Convolution Neural Networks (CNN) based time series anomaly detection to detect the additive outliers in the behavior patterns based on their spatial-temporal behavior features. We find that using Multistate LSTM is better than basic single state LSTM. The proposed method with Multistate LSTM can successfully detect the insider threats providing the AUC of 0.9042 on train data and AUC of 0.9047 on test data when trained with publically available dataset for insider threats.
Research on keystroke dynamics has the good potential to offer continuous authentication that complements conventional authentication methods in combating insider threats and identity theft before more harm can be done to the genuine users. Unfortunately, the large amount of data required by free-text keystroke authentication often contain personally identifiable information, or PII, and personally sensitive information, such as a user's first name and last name, username and password for an account, bank card numbers, and social security numbers. As a result, there are privacy risks associated with keystroke data that must be mitigated before they are shared with other researchers. We conduct a systematic study to remove PII's from a recent large keystroke dataset. We find substantial amounts of PII's from the dataset, including names, usernames and passwords, social security numbers, and bank card numbers, which, if leaked, may lead to various harms to the user, including personal embarrassment, blackmails, financial loss, and identity theft. We thoroughly evaluate the effectiveness of our detection program for each kind of PII. We demonstrate that our PII detection program can achieve near perfect recall at the expense of losing some useful information (lower precision). Finally, we demonstrate that the removal of PII's from the original dataset has only negligible impact on the detection error tradeoff of the free-text authentication algorithm by Gunetti and Picardi. We hope that this experience report will be useful in informing the design of privacy removal in future keystroke dynamics based user authentication systems.
Several recent works have addressed the ability of deep learning to disclose rich, hierarchical and discriminative models for the most diverse purposes. Specifically in the super-resolution field, Convolutional Neural Networks (CNNs) using different deep learning approaches attempt to recover realistic texture and fine grained details from low resolution images. In this work we explore the viability of these approaches for iris Super-Resolution (SR) in an iris recognition environment. For this, we test different architectures with and without a so called image re-projection to reduce artifacts applying it to different iris databases to verify the viability of the different CNNs for iris super-resolution. Results show that CNNs and image re-projection can improve the results specially for the accuracy of recognition systems using a complete different training database performing the transfer learning successfully.
Inter-modality face recognition refers to the matching of face images between different modalities and is done usually by taking visual images as source and one of the other modalities as a target. Performing facial recognition between thermal to visual is a tough task because of nonlinear spectral characteristics of thermal and visual images. However, this is a desirable requirement for night-time security applications and military surveillance. In this paper, we propose a method to improve the thermal classifier accuracy by using transfer learning and as a result, the accuracy of thermal to visual face recognition gets increased. The proposed method is tested on RGB-D-T dataset (45900 images) and UND-Xl collection (4584 images). Experimental results show that the overall accuracy of thermal to visual face recognition by transferring the knowledge gets increased to 94.32% from 89.3% on RGB-D-T dataset and from 81.54% to 90.33% on UND-Xl dataset.
The need to protect biometric data within iris systems has resulted in a number of template protection schemes. A primary issue with current template protection schemes for iris recognition is the unavoidable biometric error rates, i.e., for any given False Non-Match Rate (FNMR) there is a high False Match Rate (FMR), especially at lower values of FNMR. In this work, we primarily focus on addressing this problem using a new approach with Cuckoo Filtering simultaneously using both stable bits and discriminative bits to derive a stronger template protection scheme. The proposed template protection scheme performs in a robust manner for various configurations as compared to earlier template protection schemes that need empirical fine-tuning. With the set of experiments on a publicly available iris dataset, we benchmark our results against the stateof-art template protection scheme based on Bloom-Filters. Specifically, we demonstrate the gain in performance and robustness of proposed approach at lower FNMR and invariance of performance to configurations of template protection scheme. With a specific configuration of proposed approach, we achieve Genuine Match Rate (GMR) = 100% at FMR = 0.01% and EER = 0% in the best case and GMR = 98.44% at FMR = 0.01% and EER = 0.33% in the worst case on IITD Iris database.
If multiple face images are available for the creation of person's biometric template, some averaging method could be used to combine the feature vectors extracted from each image into a single template feature vector. Resulting average feature vector does not retain the information about image feature vector distribution. In this paper we consider the augmentation of such templates by the information about diversity of constituent face images, e.g. sample standard deviation of image feature vectors. We consider the theoretical model describing the conditions of the usefulness of template diversity measure, and see if such conditions hold in real life templates. We perform our experiments using IARPA face image datasets and deep CNN face recognizers.
Copy-move forgery is one of the most regarded image forgery technique to tamper information conveyed by the image. In this technique, segment of original image is replicated and pasted across the same image to produce forged image. This technique is capable to hide selective information or to add fictitious details in image. Detection of this form of forgery is one of the significant area of information security. In this paper, we propose block-based approach for copy-move image forgery detection to secure information conveyed through the image by identifying the forged images and to prevent spreading of tampered subject matter. Proposed model divides suspicious image in overlapping blocks. We extracted block features using Histogram of Oriented Gradients (HOG) and Singular Value Decomposition (SVD). Lexicographical sorting is performed over feature matrix followed by Euclidean distance computation to recognize similar feature vectors. To remove false match detection, Connected component labeling is utilized. Our scheme achieves highest F-measure than former techniques, when forged image sustain plain multiple copy-move, multiple copy-move with contrast adjustment, color reduction, and image blurring attacks.
Forgery activity in legal handwritten documents is an identifiable problem. Falsification of document due to minute alteration of existings not only causes immense financial loss to a person or to any organization but also lessens the economic growth of a country. Here, we introduce and present a solution to detect forgery in handwritten documents by analyzing perceptually similar ink of different pens. The research is all about forensic investigation of handwritten word alteration which is performed by adding extra letter in a way such that the whole meaning of the word changes. The problem is formulated as binary classification problem. If words of the corresponding document are written by same pen, these are classified as positive class and words of a document accompanied with little inclusion of letters as a forgery attack, are classified as negative class. The article proposes Multilayer Perceptron classifier which has been adopted to classify data instances that have been computed by extracting Y CbCr color-based statistical features. This proposal has been tested on data set which has been generated by 10 blue and 10 black ball point pens. The respective obtained average accuracy is 83.71% and 78. 18% for blue pen data and black pen data.
The widespread use of face recognition algorithms, especially in Automatic Border Control (ABC) systems has raised concerns due to potential attacks. Face morphing combines more than one face images to generate a single image that can be used in the passport enrolment procedure. Such morphed passports have proven to be a significant threat to national security, as two or more individuals that contributed to the morphed reference image can use that single travel document. In this work, we present a novel method based on hybrid colour features to automatically detect morphed face images. The proposed method is based on exploring multiple colour spaces and scale-spaces using a Laplacian pyramid to extract robust features. The texture features corresponding to each scale-space in different color spaces are extracted with Local Binary Patterns (LBP) and classified using a Spectral Regression Kernel Discriminant Analysis (SRKDA) classifier. The scores are further fused using sum rule to detect the morphed face images. Experiments are carried out on a large-scale morphed face image database consisting of printed and scanned images to reflect the real-life passport issuance scenario. The evaluation database consists of images comprised of 1270 bona fide face images and 2515 morphed face images. The performance of the proposed method is compared with seven different deep learning and seven different non-deep learning methods, which has indicated the best performance of the proposed scheme with Bona fide Presentation Classification Error (BPCER) = 0.86% @ Attack Presentation Classification Error Rate (APCER) = 10% and BPCER = 7.59% @ APCER = 5%. The obtained results indicate the robustness in detecting morphing attacks as compared to earlier works.
Mobile user authentication is a challenging task in the mobile cloud computing (MCC). In 2015, Tsai and Lo's developed authentication protocol in distributed MCC. Which is vulnerable to the biometric misuse, incorrect login credentials (password and fingerprint) and attacks for service provider impersonation. It has no provision for smart-card revocation and lacks mutual authentication. To address this-mentioned issues, we propose a novel Biometric based User Authentication Protocol for MCC. The proposed protocol supports session key agreement of participants and flawless mutual authentication. Our protocol is verified using Burrows-Abadi-Needham (BAN) logic. It further withstands all known attacks and performs well with respect to computational cost.
Online Signature Verification (OSV) is a pattern recognition problem, which involves analysis of discrete-time signals of signature samples to classify them as genuine or forgery. One of the core difficulties in designing online signature verification (OSV) system is the inherent intra-writer variability in genuine handwritten signatures, combined with the likelihood of close resemblances and dissimilarities of skilled forgeries with the genuine signatures. To address this issue, in this manuscript, we emphasize the concept of writer dependent parameter fixation (i.e. features, decision threshold and feature dimension) using interval valued representation grounded on feature fusion. For an individual writer, a subset of discriminative features is selected from the original set of features using feature clustering techniques. This is at variance with the writer independent models in which common features are used for all the writers. To practically exhibit the efficiency of the proposed model, thorough experiments are carried out on benchmarking online signature datasets MCYT-100 (DB1), MCYT-330 (DB2) consist of signatures of 100, 330 individuals respectively. Experimental result confirms the efficiency of writer specific parameters for online signature verification. The EER value, the model computes, is lower compared to various latest signature verification models.
Automatic speaker verification systems have been widely employed in a variety of commercial applications. However, advancements in the field of speech technology have equipped the attackers with sophisticated techniques for circumventing speaker verification systems. The state-of-the-art countermeasures are fairly successful in detecting speech synthesis and voice conversion attacks. However, the problem of replay attack detection has not received much attention from the researchers. In this study, we perform subband analysis on constant-Q cepstral coefficient (CQCC) and mel-frequency cepstral coefficient (MFCC) features to improve the performance of replay attack detection. We have performed experiments on the ASVspoof 2017 database which consists of 3566 genuine and 15380 replay utterances. Our experimental results suggest that the features extracted from the high frequency band carries significant discriminatory information for replay attack detection. In particular, our approach achieves an improvement of 36.33% over the baseline replay attack detection method in terms of equal error rate.
Automated fingerprint recognition using partial and latent fingerprints employs level 3 features which provide additional information in the absence of sufficient number of level 1 and level 2 features. In this paper, we present a methodology for detecting two level 3 features namely, dots and incipient ridges. Specifically, we have designed a deep convolutional neural network which generates a dot map from the input fingerprint image. Subsequently, post-processing operations are performed on the obtained dot map to identify the coordinates of dots and incipient ridges. The results of our experiments on the publicly available PolyU HRF database demonstrate the effectiveness of the proposed algorithm.
Present world has already been consistently exploring the fine edges of online and digital world by imposing multiple challenging problems/scenarios. Similar to physical world, personal identity management is very crucial inorder to provide any secure online system. Last decade has seen a lot of work in this area using biometrics such as face, fingerprint, iris etc. Still there exist several vulnerabilities and one should have to address the problem of compromised biometrics much more seriously, since they cannot be modified easily once compromised. In this work, we have proposed a secure cancelable finger dorsal template generation network (learning domain specific features) secured via. Bio-Hashing. Proposed system effectively protects the original finger dorsal images by withdrawing compromised template and reassigning the new one. A novel Finger-Dorsal Feature Extraction Net (FDFNet) has been proposed for extracting the discriminative features. This network is exclusively trained on trait specific features without using any kind of pre-trained architecture. Later Bio-Hashing, a technique based on assigning a tokenized random number to each user, has been used to hash the features extracted from FDFNet. To test the performance of the proposed architecture, we have tested it over two benchmark public finger knuckle datasets: PolyU FKP and PolyU Contactless FKI. The experimental results shows the effectiveness of the proposed system in terms of security and accuracy.
Designing an end-to-end deep learning network to match the biometric features with limited training samples is an extremely challenging task. To address this problem, we propose a new way to design an end-to-end deep CNN framework i.e., PVSNet that works in two major steps: first, an encoder-decoder network is used to learn generative domain-specific features followed by a Siamese network in which convolutional layers are pre-trained in an unsupervised fashion as an autoencoder. The proposed model is trained via triplet loss function that is adjusted for learning feature embeddings in a way that minimizes the distance between embedding-pairs from the same subject and maximizes the distance with those from different subjects, with a margin. In particular, a triplet Siamese matching network using an adaptive margin based hard negative mining has been suggested. The hyper-parameters associated with the training strategy, like the adaptive margin, have been tuned to make the learning more effective on biometric datasets. In extensive experimentation, the proposed network outperforms most of the existing deep learning solutions on three type of typical vein datasets which clearly demonstrates the effectiveness of our proposed method.
At present spoofing attacks via which biometric system is potentially vulnerable against a fake biometric characteristic, introduces a great challenge to recognition performance. Despite the availability of a broad range of presentation attack detection (PAD) or liveness detection algorithms, fingerprint sensors are vulnerable to spoofing via fake fingers. In such situations, finger dorsal images can be thought of as an alternative which can be captured without much user cooperation and are more appropriate for outdoor security applications. In this paper we present a first feasibility study of spoofing attack scenarios on finger dorsal authentication system, which include four types of presentation attacks such as printed paper, wrapped printed paper, scan and mobile. This study also presents a CNN based spoofing attack detection method which employ state-of-the-art deep learning techniques along with transfer learning mechanism. We have collected 196 finger dorsal real images from 33 subjects, captured with a Lytro camera and also created a set of 784 finger dorsal spoofing images. Extensive experimental results have been performed that demonstrates the superiority of the proposed approach for various spoofing attacks.
The principle of Photo Response Non-Uniformity (PRNU) is used to link an image with its source, i.e., the sensor that produced it. In this work, we investigate if it is possible to modify an iris image acquired using one sensor in order to spoof the PRNU noise pattern of a different sensor. In this regard, we develop an image perturbation routine that iteratively modifies blocks of pixels in the original iris image such that its PRNU pattern approaches that of a target sensor. Experiments indicate the efficacy of the proposed perturbation method in spoofing PRNU patterns present in an iris image whilst still retaining its biometric content.
Threats from the inside of an organization's perimeters are a significant problem since it is difficult to distinguish them from benign activities. Recent reports indicate that the accidental/unintentional incidents account for the majority ofall insider security incidents. Human error is a major factor in unintentional insider threat. In this paper, we propose a novel approach for unintentional insider threat (UIT) detection and mitigation based on eye movement patterns. We perform experiments to capture unique characteristics of a user's eye movements as they perform several computer-based activities in different scenarios. The goal is to evaluate the effectiveness of using eye movement patterns in determining a user's subjective mental workload which is one of the main contributing factors to human error. We extract eye movement and pupil features which allow us to reliably achieve this goal. We evaluate our proposed approach using several classifiers and examine how different subsets of features affect the performance. The results show about 82% accuracy on average for users wearing eye glasses and an average accuracy of 84.5% for users without eye glasses. Our results demonstrate that users' eye movement patterns and pupil behaviors can reveal valuable clues about their subjective mental workload and could be used in developing effective tools for unintentional insider threat detection and mitigation in real-world environments.
Biometrie identification systems have become immensely popular and important because of their high reliability and efficiency. However person identification at a distance, still remains a challenging problem. Gait can be seen as an essential biometric feature for human recognition and identification. It can be easily acquired from a distance and does not require any user cooperation thus making it suitable for surveillance. But the task of recognizing an individual using gait can be adversely affected by varying view points making this task more and more challenging. Our proposed approach tackles this problem by identifying spatio-temporal features and performing extensive experimentation and training mechanism. In this paper, we propose a 3-D Convolution Deep Neural Network for person identification using gait under multiple view. It is a 2-stage network, in which we have a classification network that initially identifies the viewing point angle. After that another set of networks (one for each angle) has been trained to identify the person under a particular viewing angle. We have tested this network over CASIA-B publicly available database and have achieved state-of-the-art results. The proposed system is much more efficient in terms of time and space and performing better for almost all angles.
