Signature Representations Research Articles

TransOSV: Offline Signature Verification with Transformers

Signature verification is a frequently-used forensics technology in numerous safety-critical situations. Although convolutional neural networks (CNNs) have made significant advancements in the field of signature verification, their reliance on local neighborhood operations poses limitations in capturing the global contextual relationships among signature strokes. To overcome this weakness, in this paper, we propose a novel holistic-part unified model named TransOSV based on the vision transformer framework to solve offline signature verification problem. The signature images are first encoded into patch sequences by the proposed transformer-based holistic encoder to learn the global signature representation. Second, considering the subtle local difference between the genuine signature and forged signature, we design a contrast based part decoder along with a sparsity loss, which are utilized to learn the discriminative part features. With the learned holistic features and part features, the proposed model is optimized by the contrast loss function. To reduce the influence of sample imbalance, we also formulate a new focal contrast loss function. Furthermore, we conduct the proposed model to learn signature representations for writer-dependent signature verification task. The experimental results demonstrate the potential of the proposed TransOSV model for both writer-independent and writer-dependent signature verification tasks, achieving remarkable performance improvements and competitive results on four widely-used offline signature datasets.

SynSig2Vec: Forgery-Free Learning of Dynamic Signature Representations by Sigma Lognormal-Based Synthesis and 1D CNN.

Handwritten signature verification is a challenging task because signatures of a writer may be skillfully imitated by a forger. As skilled forgeries are generally difficult to acquire for training, in this paper, we propose a deep learning-based dynamic signature verification framework, SynSig2Vec, to address the skilled forgery attack without training with any skilled forgeries. Specifically, SynSig2Vec consists of a novel learning-by-synthesis method for training and a 1D convolutional neural network model, called Sig2Vec, for signature representation extraction. The learning-by-synthesis method first applies the Sigma Lognormal model to synthesize signatures with different distortion levels for genuine template signatures, and then learns to rank these synthesized samples in a learnable representation space based on average precision optimization. The representation space is achieved by the proposed Sig2Vec model, which is designed to extract fixed-length representations from dynamic signatures of arbitrary lengths. Through this training method, the Sig2Vec model can extract extremely effective signature representations for verification. Our SynSig2Vec framework requires only genuine signatures for training, yet achieves state-of-the-art performance on the largest dynamic signature database to date, DeepSignDB, in both skilled forgery and random forgery scenarios. Source codes of SynSig2Vec will be available at https://github.com/LaiSongxuan/SynSig2Vec.

Discussion of an overview of some classical models and discussion of the signature‐based models of preventive maintenance

Discussion of an overview of some classical models and discussion of the signature‐based models of preventive maintenance

R-SigNet: Reduced space writer-independent feature learning for offline writer-dependent signature verification

Handwritten signatures are a widespread biometric trait for person identification and verification. Reliable authentication and authorization of individuals are, however, challenging tasks due to possible skilled forgeries; especially when a forger has access to a given signature and deliberately tries to imitate it. This problem is even more emphasised in offline signature verification, where dynamic signature information is lost, resulting, as a consequence, in an increased difficulty discerning between genuine and forged signatures. To address this issue, solutions based on convolutional neural networks (CNN) are currently being devised to automatically extract features from a signature. Although highly performing, these methods require a high number of learnable parameters to produce meaningful signature representations, ultimately leading to long training times. In this paper, the R-SigNet architecture, a multi-task approach exploiting a relaxed loss to learn a reduced feature space for writer-independent (WI) signature verification, is presented. Compact generic features are automatically extracted by this network, so that a support vector machine (SVM) can be trained and tested in offline writer-dependent (WD) mode. By leveraging a small generic feature space, the proposed system achieves improved performances and reduced training times with respect to the current literature, as shown by the results obtained on several benchmark datasets.

A Note on the Signature Representations of the Symmetric Groups

For a partition [Formula: see text] and a prime [Formula: see text], we prove a necessary and sufficient condition for there to exist a composition [Formula: see text] such that [Formula: see text] can be obtained from [Formula: see text] after rearrangement and no partial sums of [Formula: see text] are divisible by [Formula: see text]. To demonstrate why we are interested in the question, we compute some signed [Formula: see text]-Kostka numbers.

Conditions on marginals and copula of component lifetimes for signature representation of system lifetime

The signature of a system is a probability vector that depends only on the system structure. Under the classic IID (independent and identically distributed) assumption on the component lifetimes, the system lifetime distribution is the convex combination of consecutive component failure times, and the signature coordinates constitute the mixture coefficients. In this case the signature representations are very useful in determining the system lifetime distributions and for stochastic comparisons of them. This first representation was obtained in 1985 by Samaniego. Then it was extended to the more general case of exchangeable component lifetimes. In 2011 Marichal, Mathonet and Waldhauser presented necessary and sufficient conditions assuring the Samaniego representation. There were expressed in terms of distributional properties of families of auxiliary indicator random vectors parametrized by positive numbers. In the paper we obtain other necessary and sufficient conditions represented in terms of the marginal distributions of component lifetimes and the dependence copula of them. Moreover, we study symmetry conditions for the equality of structural and probabilistic signatures.

Comparative genomic signature representations of the emerging COVID-19 coronavirus and other coronaviruses: High identity and possible recombination between Bat and Pangolin coronaviruses

Coronaviruses are responsible on respiratory diseases in animal and human. The combination of numerical encoding techniques and digital signal processing methods are becoming increasingly important in handling large genomic data. In this paper, we propose to analyze the SARS-CoV-2 genomic signature using the combination of different nucleotide representations and signal processing tools in the aim to identify its genetic origin. The sequence of SARS-CoV-2 was compared with 21 relevant sequences including Bat, Yak and Pangolin coronavirus sequences. In addition, we developed a new algorithm to locate the nucleotide modifications. The results show that the Bat and Pangolin coronaviruses were the most related to SARS-CoV-2 with 96% and 86% of identity all along the genome. Within the S gene sequence, the Pangolin sequence presents local highest nucleotide identity. Those findings suggest genesis of SARS-Cov-2 through evolution from Bat and Pangolin strains. This study offers new ways to automatically characterize viruses.

SynSig2Vec: Learning Representations from Synthetic Dynamic Signatures for Real-World Verification

An open research problem in automatic signature verification is the skilled forgery attacks. However, the skilled forgeries are very difficult to acquire for representation learning. To tackle this issue, this paper proposes to learn dynamic signature representations through ranking synthesized signatures. First, a neuromotor inspired signature synthesis method is proposed to synthesize signatures with different distortion levels for any template signature. Then, given the templates, we construct a lightweight one-dimensional convolutional network to learn to rank the synthesized samples, and directly optimize the average precision of the ranking to exploit relative and fine-grained signature similarities. Finally, after training, fixed-length representations can be extracted from dynamic signatures of variable lengths for verification. One highlight of our method is that it requires neither skilled nor random forgeries for training, yet it surpasses the state-of-the-art by a large margin on two public benchmarks.

Variable star classification using multiview metric learning

ABSTRACTComprehensive observations of variable stars can include time domain photometry in a multitude of filters, spectroscopy, estimates of colour (e.g. U-B), etc. When the objective is to classify variable stars, traditional machine learning techniques distill these various representations (or views) into a single feature vector and attempt to discriminate among desired categories. In this work, we propose an alternative approach that inherently leverages multiple views of the same variable star. Our multiview metric learning framework enables robust characterization of star categories by directly learning to discriminate in a multifaceted feature space, thus, eliminating the need to combine feature representations prior to fitting the machine learning model. We also demonstrate how to extend standard multiview learning, which employs multiple vectorized views, to the matrix-variate case which allows very novel variable star signature representations. The performance of our proposed methods is evaluated on the UCR Starlight and LINEAR data sets. Both the vector and matrix-variate versions of our multiview learning framework perform favourably – demonstrating the ability to discriminate variable star categories.

Learning Global-Local Distance Metrics for Signature-Based Biometric Cryptosystems

Biometric traits, such as fingerprints, faces and signatures have been employed in bio-cryptosystems to secure cryptographic keys within digital security schemes. Reliable implementations of these systems employ error correction codes formulated as simple distance thresholds, although they may not effectively model the complex variability of behavioral biometrics like signatures. In this paper, a Global-Local Distance Metric (GLDM) framework is proposed to learn cost-effective distance metrics, which reduce within-class variability and augment between-class variability, so that simple error correction thresholds of bio-cryptosystems provide high classification accuracy. First, a large number of samples from a development dataset are used to train a global distance metric that differentiates within-class from between-class samples of the population. Then, once user-specific samples are available for enrollment, the global metric is tuned to a local user-specific one. Proof-of-concept experiments on two reference offline signature databases confirm the viability of the proposed approach. Distance metrics are produced based on concise signature representations consisting of about 20 features and a single prototype. A signature-based bio-cryptosystem is designed using the produced metrics and has shown average classification error rates of about 7% and 17% for the PUCPR and the GPDS-300 databases, respectively. This level of performance is comparable to that obtained with complex state-of-the-art classifiers.

Spectral Unmixing-Based Clustering of High-Spatial Resolution Hyperspectral Imagery

This paper introduces a novel unsupervised spectral unmixing-based clustering method for high-spatial resolution hyperspectral images ( HSIs ). In contrast to most clustering methods reported so far, which are applied on the spectral signature representations of the image pixels, the idea in the proposed method is to apply clustering on the abundance representations of the pixels. Specifically, the proposed method comprises two main processing stages namely: an unmixing stage (consisting of the endmember extraction and abundance estimation (AE) substages) and a clustering stage. In the former stage, suitable endmembers are selected first as the most representative pure pixels. Then, the spectral signature of each pixel is expressed as a linear combination of the endmembers’ spectral signatures and the pixel itself is represented by the relative abundance vector, which is estimated via an efficient AE algorithm. The resulting abundance vectors associated with the HSI pixels are next fed to the clustering stage. Eventually, the pixels are grouped into clusters, in terms of their associated abundance vectors and not their spectral signatures. Experiments are performed on a synthetic HSI dataset as well as on three airborne HSI datasets of high-spatial resolution containing vegetation and urban areas. The experimental results corroborate the effectiveness of the proposed method and demonstrate that it outperforms state-of-the-art clustering techniques in terms of overall accuracy, average accuracy, and kappa coefficient.

Image retrieval by information fusion based on scalable vocabulary tree and robust Hausdorff distance

In recent years, Scalable Vocabulary Tree (SVT) has been shown to be effective in image retrieval. However, for general images where the foreground is the object to be recognized while the background is cluttered, the performance of the current SVT framework is restricted. In this paper, a new image retrieval framework that incorporates a robust distance metric and information fusion is proposed, which improves the retrieval performance relative to the baseline SVT approach. First, the visual words that represent the background are diminished by using a robust Hausdorff distance between different images. Second, image matching results based on three image signature representations are fused, which enhances the retrieval precision. We conducted intensive experiments on small-scale to large-scale image datasets: Corel-9, Corel-48, and PKU-198, where the proposed Hausdorff metric and information fusion outperforms the state-of-the-art methods by about 13, 15, and 15%, respectively.

The importance of topography-controlled sub-grid process heterogeneity and semi-quantitative prior constraints in distributed hydrological models

Abstract. Heterogeneity of landscape features like terrain, soil, and vegetation properties affects the partitioning of water and energy. However, it remains unclear to what extent an explicit representation of this heterogeneity at the sub-grid scale of distributed hydrological models can improve the hydrological consistency and the robustness of such models. In this study, hydrological process complexity arising from sub-grid topography heterogeneity was incorporated into the distributed mesoscale Hydrologic Model (mHM). Seven study catchments across Europe were used to test whether (1) the incorporation of additional sub-grid variability on the basis of landscape-derived response units improves model internal dynamics, (2) the application of semi-quantitative, expert-knowledge-based model constraints reduces model uncertainty, and whether (3) the combined use of sub-grid response units and model constraints improves the spatial transferability of the model. Unconstrained and constrained versions of both the original mHM and mHMtopo, which allows for topography-based sub-grid heterogeneity, were calibrated for each catchment individually following a multi-objective calibration strategy. In addition, four of the study catchments were simultaneously calibrated and their feasible parameter sets were transferred to the remaining three receiver catchments. In a post-calibration evaluation procedure the probabilities of model and transferability improvement, when accounting for sub-grid variability and/or applying expert-knowledge-based model constraints, were assessed on the basis of a set of hydrological signatures. In terms of the Euclidian distance to the optimal model, used as an overall measure of model performance with respect to the individual signatures, the model improvement achieved by introducing sub-grid heterogeneity to mHM in mHMtopo was on average 13 %. The addition of semi-quantitative constraints to mHM and mHMtopo resulted in improvements of 13 and 19 %, respectively, compared to the base case of the unconstrained mHM. Most significant improvements in signature representations were, in particular, achieved for low flow statistics. The application of prior semi-quantitative constraints further improved the partitioning between runoff and evaporative fluxes. In addition, it was shown that suitable semi-quantitative prior constraints in combination with the transfer-function-based regularization approach of mHM can be beneficial for spatial model transferability as the Euclidian distances for the signatures improved on average by 2 %. The effect of semi-quantitative prior constraints combined with topography-guided sub-grid heterogeneity on transferability showed a more variable picture of improvements and deteriorations, but most improvements were observed for low flow statistics.

Phylogenetic Signals in DNA Composition: Limitations and Prospects

The concept of genome signature allows sequence comparisons without alignment. It relies on the premise that oligonucleotide compositions of DNA segments from the same or closely related genomes tend to be more similar than those from distantly related genomes. This concept has been used in detection of lateral gene transfer, phylogenetic classification of metagenome sequences (binning), and in studies of evolution of viruses and plasmids. The goal of this work is to explore limitations of genome signature in phylogenetic classification of DNA sequences and to identify formal representations of genome signature that expose best the phylogenetic relationships among prokaryotes. We found that genome signatures that best represent phylogenetic relationships are those normalized to factor out differences in G + C content and utilizing the standard A-C-G-T alphabet or the degenerate R-Y (purine-pyrimidine) alphabet. The main limitation of all genome signature representations tested is lack of divergence among some distantly related species. "Crowding" of the genome signature space and absence of molecular clock likely contribute to this phenomenon. We introduce "periodicity signatures"--formal representations of periodic sequence patterns related to DNA curvature--which can discriminate between bacterial and archaeal DNA sequences. Interestingly, archaea of the order Halobacteriaceae have periodic signatures similar to bacteria, possibly due to their early divergence from other archaea, extensive lateral gene transfer, or due to their adaptation to high salt environments. Our results have practical implications for development and application of genome signature-based methods for analysis and classification of DNA sequences.

A novel bioinformatic strategy for unveiling hidden genome signatures of eukaryotes: self-organizing map of oligonucleotide frequency.

With the increasing amount of available genome sequences, novel tools are needed for comprehensive analysis of species-specific sequence characteristics for a wide variety of genomes. We used an unsupervised neural network algorithm, Kohonen's self-organizing map (SOM), to analyze di- and trinucleotide frequencies in 9 eukaryotic genomes of known sequences (a total of 1.2 Gb); S. cerevisiae, S. pombe, C. elegans, A. thaliana, D. melanogaster, Fugu, and rice, as well as P. falciparum chromosomes 2 and 3, and human chromosomes 14, 20, 21, and 22, that have been almost completely sequenced. Each genomic sequence with different window sizes was encoded as a 16- and 64-dimensional vector giving relative frequencies of di- and trinucleotides, respectively. From analysis of a total of 120,000 nonoverlapping 10-kb sequences and overlapping 100-kb sequences with a moving step size of 10 kb, derived from a total of the 1.2 Gb genomic sequences, clear species-specific separations of most sequences were obtained with the SOMs. The unsupervised algorithm could recognize, in most of the 120,000 10-kb sequences, the species-specific characteristics (key combinations of oligonucleotide frequencies) that are signature representations of each genome. Because the classification power is very high, the SOMs can provide fundamental bioinformatic strategies for extracting a wide range of genomic information that could not otherwise be obtained.