Modern neural recording techniques such as two-photon imaging or Neuropixel probes allow to acquire vast time-series datasets with responses of hundreds or thousands of neurons. Contrastive learning is a powerful self-supervised framework for learning representations of complex datasets. Existing applications for neural time series rely on generic data augmentations and do not exploit the multi-trial data structure inherent in many neural datasets. Here we present TRACE, a new contrastive learning framework that averages across different subsets of trials to generate positive pairs. TRACE allows to directly learn a two-dimensional embedding, combining ideas from contrastive learning and neighbor embeddings. We show that TRACE outperforms other methods, resolving fine response differences in simulated data. Further, using in vivo recordings, we show that the representations learned by TRACE capture both biologically relevant continuous variation, cell-type-related cluster structure, and can assist data quality control.

Criminal investigators, faced with the complexity of knowledge discovery in major cases, can benefit from integrating information retrieval and natural language processing in their workflow. We explore the application of pre-trained sentence embeddings for clustering and semantic search, presenting promising results. Evaluating over 3,700 Norwegian police reports from high-profile cases; we find that sentence embeddings enhance information retrieval speed and reveal previously unnoticed data, offering valuable insights for future applications in the field of criminal investigations.

In recent years, most exciting inputs (MEIs) synthesized from encoding models of neuronal activity have become an established method for studying tuning properties of biological and artificial visual systems. However, as we move up the visual hierarchy, the complexity of neuronal computations increases. Consequently, it becomes more challenging to model neuronal activity, requiring more complex models. In this study, we introduce a novel readout architecture inspired by the mechanism of visual attention. This new architecture, which we call attention readout, together with a data-driven convolutional core outperforms previous task-driven models in predicting the activity of neurons in macaque area V4. However, as our predictive network becomes deeper and more complex, synthesizing MEIs via straightforward gradient ascent (GA) can struggle to produce qualitatively good results and overfit to idiosyncrasies of a more complex model, potentially decreasing the MEI's model-to-brain transferability. To solve this problem, we propose a diffusion-based method for generating MEIs via Energy Guidance (EGG). We show that for models of macaque V4, EGG generates single neuron MEIs that generalize better across varying model architectures than the state-of-the-art GA, while at the same time reducing computational costs by a factor of 4.7x, facilitating experimentally challenging closed-loop experiments. Furthermore, EGG diffusion can be used to generate other neurally exciting images, like most exciting naturalistic images that are on par with a selection of highly activating natural images, or image reconstructions that generalize better across architectures. Finally, EGG is simple to implement, requires no retraining of the diffusion model, and can easily be generalized to provide other characterizations of the visual system, such as invariances. Thus, EGG provides a general and flexible framework to study the coding properties of the visual system in the context of natural images.

Reverse engineering of file systems is indispensable for tool testing, accurate evidence acquisition, and correct interpretation of data structures by law enforcement in criminal investigations. This position paper examines emerging techno-legal challenges from the practice of reverse engineering for law enforcement purposes. We demonstrate that this new context creates uncertainties about the legality of tools and methods used for evidence acquisition and the compliance of law enforcement with obligations to protect intellectual property and confidential information. Further identified are gaps between legal provisions and practice related to disclosure and peer-review of sensitive digital forensic methodology, trade secrets in investigations, and governmental vulnerability disclosure. It is demonstrated that reverse engineering of file systems is insufficiently addressed by legislators, which results in a lack of file system interpretation and validation information for law enforcement and their dependence on tools. Outlined are recommendations for further developments of digital forensic regulation.

Integrating data from multiple experiments is common practice in systems neuroscience but it requires inter-experimental variability to be negligible compared to the biological signal of interest. This requirement is rarely fulfilled; systematic changes between experiments can drastically affect the outcome of complex analysis pipelines. Modern machine learning approaches designed to adapt models across multiple data domains offer flexible ways of removing inter-experimental variability where classical statistical methods often fail. While applications of these methods have been mostly limited to single-cell genomics, in this work, we develop a theoretical framework for domain adaptation in systems neuroscience. We implement this in an adversarial optimization scheme that removes inter-experimental variability while preserving the biological signal. We compare our method to previous approaches on a large-scale dataset of two-photon imaging recordings of retinal bipolar cell responses to visual stimuli. This dataset provides a unique benchmark as it contains biological signal from well-defined cell types that is obscured by large inter-experimental variability. In a supervised setting, we compare the generalization performance of cell type classifiers across experiments, which we validate with anatomical cell type distributions from electron microscopy data. In an unsupervised setting, we remove inter-experimental variability from the data which can then be fed into arbitrary downstream analyses. In both settings, we find that our method achieves the best trade-off between removing inter-experimental variability and preserving biological signal. Thus, we offer a flexible approach to remove inter-experimental variability and integrate datasets across experiments in systems neuroscience. Code available at https://github.com/eulerlab/rave.

Cyber threat intelligence (CTI) can be gathered from multiple sources, and Twitter is one such open source platform where a large volume and variety of threat data is shared every day. The automated and timely mining of relevant threat knowledge from this data can be crucial for enrichment of existing threat intelligence platforms to proactively defend against cyber attacks. We propose CTI-Twitter: a novel frame-work combining supervised and unsupervised learning models to collect, process, analyze and generate threat specific knowledge from tweets coming from multiple users. CTI-Twitter has multi-fold contributions: i) first collecting tweets through Twitter API, ii) extracting relevant threat tweets from irrelevant ones, and classifying relevant ones into multiple classes of threats iii) then grouping tweets belonging to each class using topic modeling iv) finally performing data enrichment and verification process. We evaluate our proposed model on real-time tweets collected for about four months (in year 2020) using Twitter API. The encouraging results obtained indicate the effectiveness of CTI-Twitter in terms of timeliness and discovery of trending attacks patterns, and vulnerabilities.

Web applications are often exploited using different techniques like injection, buffer overflow, etc. An HTTP request carrying such malicious content will be different from a normal request. In this paper we propose to detect such anomalous HTTP requests using regular expression based signatures. These signatures are generated using character combinations specifically identified from known malicious requests. We identify certain characters which are useful for differentiating normal and anomalous requests using their frequency value comparison and subsequently select those combinations which have high chances of appearing together by estimating their joint probability values. We experiment with few sample attack types and show that proposed method can identify anomalous HTTP requests.

A few highly skilled cybercriminals run the Crime as a Service business model. These expert hackers provide entry-level criminals with tools that allow them to enhance their cybercrime operations significantly. Thus, effectively and efficiently disrupting highly proficient cybercriminals is of a high priority to law enforcement. Such individuals can be found in vast underground forums, though it is particularly challenging to identify and profile individual users. We tackle this problem by combining two analysis methods: text analysis with Latent Dirichlet Allocation (LDA) and Social Network Analysis with centrality measures. In this paper, we use LDA to eliminate around 79% of hacker forum users with very low to no technical skills, while also inferring the forum roles held by the remaining users. Furthermore, we use centrality measures to identify users with hugely popular public posts, including users with very few public posts who receive much attention from their peers. We study various preprocessing methods, wherein we achieve our results by following a series of rigorous preprocessing steps. Our proposed method works towards overcoming current challenges in identifying and interrupting highly proficient cybercriminals.

Visual processing in the retina has been studied in great detail at all levels such that a comprehensive picture of the retina's cell types and the many neural circuits they form is emerging. However, the currently best performing models of retinal function are black-box CNN models which are agnostic to such biological knowledge. In particular, these models typically neglect the role of the many inhibitory circuits involving amacrine cells and the biophysical mechanisms underlying synaptic release. Here, we present a computational model of temporal processing in the inner retina, including inhibitory feedback circuits and realistic synaptic release mechanisms. Fit to the responses of bipolar cells, the model generalized well to new stimuli including natural movie sequences, performing on par with or better than a benchmark black-box model. In pharmacology experiments, the model replicated in silico the effect of blocking specific amacrine cell populations with high fidelity, indicating that it had learned key circuit functions. Also, more in depth comparisons showed that connectivity patterns learned by the model were well matched to connectivity patterns extracted from connectomics data. Thus, our model provides a biologically interpretable data-driven account of temporal processing in the inner retina, filling the gap between purely black-box and detailed biophysical modeling.

Underground forums serves as gathering place for like-minded cyber criminals and are an continued threat to law and order. Law enforcement agencies can use Open-Source Intelligence (OSINT) to gather valuable information to proactively counter existing and new threats. For example, by shifting criminal investigation's focus onto certain cyber criminals with large impact in underground forums and related criminal business models. This paper presents our study on text preprocessing requirements and document construction for the topic model algorithm Latent Dirichlet Allocation (LDA). We identify a set of preprocessing requirements based on literature review and demonstrate them on a real-world forum, similar to those used by cyber criminals. Our result show that topic modelling processes needs to follow a very strict procedure to provide significant result that can be useful in OSINT. Additionally, more reliable results are produced by tuning the hyper-parameters and the number of topics for LDA. We demonstrate improved results by iterative preprocessing to continuously improve the model, which provide more coherent and focused topics.

Variability, stochastic or otherwise, is a central feature of neural activity. Yet the means by which estimates of variation and uncertainty are derived from noisy observations of neural activity is often heuristic, with more weight given to numerical convenience than statistical rigour. For two-photon imaging data, composed of fundamentally probabilistic streams of photon detections, the problem is particularly acute. Here, we present a statistical pipeline for the inference and analysis of neural activity using Gaussian Process regression, applied to two-photon recordings of light-driven activity in ex vivo mouse retina. We demonstrate the flexibility and extensibility of these models, considering cases with non-stationary statistics, driven by complex parametric stimuli, in signal discrimination, hierarchical clustering and other inference tasks. Sparse approximation methods allow these models to be fitted rapidly, permitting them to actively guide the design of light stimulation in the midst of ongoing two-photon experiments.

[This corrects the article DOI: 10.1371/journal.pcbi.1007205.].

Traditional security controls, such as firewalls, anti-virus and IDS, are ill-equipped to help IT security and response teams keep pace with the rapid evolution of the cyber threat landscape. Cyber Threat Intelligence (CTI) can help remediate this problem by exploiting non-traditional information sources, such as hacker forums and "dark-web" social platforms. Security and response teams can use the collected intelligence to identify emerging threats. Unfortunately, when manual analysis is used to extract CTI from non-traditional sources, it is a time consuming, error-prone and resource intensive process. We address these issues by using a hybrid Machine Learning model that automatically searches through hacker forum posts, identifies the posts that are most relevant to cyber security and then clusters the relevant posts into estimations of the topics that the hackers are discussing. The first (identification) stage uses Support Vector Machines and the second (clustering) stage uses Latent Dirichlet Allocation. We tested our model, using data from an actual hacker forum, to automatically extract information about various threats such as leaked credentials, malicious proxy servers, malware that evades AV detection, etc. The results demonstrate our method is an effective means for quickly extracting relevant and actionable intelligence that can be integrated with traditional security controls to increase their effectiveness.

Hacker forums and other social platforms may contain vital information about cyber security threats. But using manual analysis to extract relevant threat information from these sources is a time consuming and error-prone process that requires a significant allocation of resources. In this paper, we explore the potential of Machine Learning methods to rapidly sift through hacker forums for relevant threat intelligence. Utilizing text data from a real hacker forum, we compared the text classification performance of Convolutional Neural Network methods against more traditional Machine Learning approaches. We found that traditional machine learning methods, such as Support Vector Machines, can yield high levels of performance that are on par with Convolutional Neural Network algorithms.

The amount of data seized in Crime Investigations has increased enormously. Investigators are more than ever confronted with vast amount of heterogeneous data, highly-diverse data formats, increased complexity in distributed stored information. With constantly increasing network bandwidth it makes extremely challenging to process or even store part of the network traffic. Nevertheless, criminal investigations need to solve crimes in a timely manners. New computational methods, infrastructure and algorithmic approaches are required. Although Big Data is a challenge for criminal investigators, it can also help them make to source an detect patterns to prevent and solve crimes. This paper aims to raise attention to current challenges in Cyber Crime Investigations - related to Big Data - and possible ways to approach combating cybercrimes.

Malware classification has become an important task in protection of privacy and sensitive information from being stolen or modified. A number of malware categories and families emerged over last decade targeting Microsoft Windows since it is the most attractive platform for virus developers. Software for this OS is provided in a format of Portable Executable (PE) files. Majority of commercial anti-virus solutions use signature-based detection, where malware pattern is described by means of unique crisp identifier of corresponding PE file content. Neuro-Fuzzy is one of the prospective Hybrid Intelligence methods suitable for malware detection. Despite the fact that Neuro-Fuzzy is being utilized successfully for binary malware classification, it was not investigated in such complex cases as multinomial classification of malware categories and families. The advantage of this method is ability to produce generalized fuzzy rules model applicable for real-world application. For this study we created a novel large-scale malware dataset that include variety of malware samples. Moreover, reports from VirusTotal and PEframe were collected to PE header features and species naming used by major anti-virus vendors. Finally, we applied tuned Neuro-Fuzzy for handling multinomial problems. This paper contributes as a stepping stone for future analysis of aspects Neuro-Fuzzy methods for real-world malware classification.

Malware detection has been widely explored in recent years due to an increased rate of information theft, ransom demands and money laundering cases. It is known that MS Windows Operating System family is susceptible to attacks due to the extensive number of found vulnerabilities across many outdated versions that are still in use. As a result, multiple malware categories and families have emerged. Many researchers stress the importance of multinomial malware classification rather than conventional malicious versus benign classification. However, multinomial detection has been neither sufficiently explored nor tested on recent malware samples. We believe that dynamic analysis can reveal information more relevant to classification characteristics in each malware category, and that cannot be done by static analysis. In this ongoing research, we explore a novel application of Machine Learning classification on a multinomial malware using behavioural analysis in a controlled environment. A novel dataset containing recent malware samples was used to show a prospective application of the automated multinomial malware detection using different artifacts left in the system. Corresponding methodology was proposed to extract characteristics from the system artifacts. We believe that this work can facilitate a decision support when it is critical to narrow down the threat vectors and to find similarity in dealing with zero-day attacks.

Abstract In the detection of web attacks, it is necessary that Web Application Firewalls (WAFs) are effective, at the same time than efficient. In this paper, we propose a new methodology for web attack detection that enhances these two aspects of WAFs. It involves both feature construction and feature selection. For the feature construction phase, many professionals rely on their expert knowledge to define a set of important features, what normally leads to high and reliable attack detection rates. Nevertheless, it is a manual process and not quickly adaptive to the changing network environments. Alternatively, automatic feature construction methods (such asn‐grams) overcome this drawback, but they provide unreliable results. Therefore, in this paper, we propose to combine expert knowledge withn‐gram feature construction method for reliable and efficient web attack detection. However, the number ofn‐grams grows exponentially withn, which usually leads to high dimensionality problems. Hence, we propose to apply feature selection to reduce the number of redundant and irrelevant features. In particular, we study the recently proposed Generic Feature Selection (GeFS) measure, which has been successfully tested in intrusion detection systems. Additionally, we use several decision tree algorithms as classifiers of WAFs. The experiments are conducted on the publicly available ECML/PKDD 2007 dataset. The results show that the combination of expert knowledge andn‐grams outperforms each separate technique and that the GeFS measure can greatly reduce the number of features, thus enhancing both the effectiveness and efficiency of WAFs. Copyright © 2012 John Wiley & Sons, Ltd.

The forensics investigation requirements are in direct conflict with the privacy rights of those whose actions are being investigated. At the same time, once the private data is exposed it is impossible to `undo' its exposure effects should the suspect is found innocent! Moreover, it is not uncommon that during a suspect investigation, private information of other innocent parties becomes apparent to the forensics investigator. These all raise the concern for development of platforms for enforcing privacy boundaries even to authorized forensics investigators. To the best of authors' knowledge, there is no practical model for privacy-respecting digital investigation which is capable of considering different jurisdictions requirements and protecting subjects' data privacy in line with investigation warrant permissions and data-origin privacy requirements. Privacy-respecting digital forensics as an emerging cross-disciplinary research area is moving toward addressing above issues. In this paper, we first establish needed foundations and describe details of "privacy-respecting digital investigation" as a cross-disciplinary field of research. Afterwards, we review main research efforts in different research disciplines relevant to the field and elaborate existing research problems. We finalize the paper by looking at potential privacy issues during digital investigation in the light of EU, US, and APEC privacy regulations. The main contributions of this paper are first establishing essential foundations and providing detailed definition of "privacy-respecting digital investigation" as a new cross-disciplinary field of research, second a review of current state of art in different disciplines relevant to this field, third elaborating existing issues and discussing most promising solutions relevant to these disciplines, and forth is detailed discussion of potential privacy issues in different phases of digital forensics life cycle based on EU,US, and APEC privacy regulations. We hope this paper opens up a new and fruitful avenue in the study, design, and development of privacy respecting forensics investigation as an interdisciplinary field of research.

Automatic identification of an individual based on his/her handwriting characteristics is an important forensic tool. In a computational forensic scenario, presence of huge amount of text/information in a questioned document cannot be ensured. Lack of data threatens system reliability in such cases. We here propose a writer identification system for Oriya script which is capable of performing reasonably well even with small amount of text. Experiments with curvature feature are reported here, using Support Vector Machine (SVM) as classifier. We got promising results of 94.00% writer identification accuracy at first top choice and 99% when considering first three top choices.

Classification/misclassification of similar shaped characters largely affects OCR accuracy. Sometimes occlusion/insertion of a part of character (due to inferior scanning quality) also makes it look alike another character type. For such adverse situations a part based character recognition system could be more effective. In order to encounter mentioned adverse scenario we propose a new feature encoding technique. This feature encoding is based on the amalgamation of Gabor filter-based features with SURF features (G-SURF). Features generated from a character are provided to Support Vector Machine (SVM) classifier. We obtained an encouraging accuracy on similar shaped characters from three different scripts.

Adaptation of Intrusion Detection Systems (IDSs) in the heterogeneous and adversarial network environments is crucial. We design an adaptive IDS that has 10% higher accuracy than the best of four different baseline IDSs. Rather than creating a new `super' IDS, we combine the outputs of the IDSs by using the online learning framework proposed by Bousquet and Warmuth [1]. The combination framework allows to dynamically determine the best IDSs performed in different segments of a dataset. Moreover, to increase the accuracy and reliability of the intrusion detection results, the fusion between outputs of the four IDSs is taken into account by a new expanded framework. We conduct the experiments on two different datasets for benchmarking Web Application Firewalls: the ECML-PKDD 2007 HTTP dataset and the CISIC HTTP 2010. Experimental results show the high adaptability of the proposed IDS.

Feature selection for botnet malware detection is an important task. In this paper, we study the recently proposed Generic-Feature-Selection (GeFS) measure [18]. Since there is no benchmark dataset of botnet malware, we conduct experiments on the dataset that is generated by using public available tools. We utilize the static and dynamic approaches [24], [29], [12] to extract features from the generated dataset and to produce two separate feature sets. We analyze the statistical properties of these feature sets to provide more insights of their nature and quality. Subsequently we determine appropriate instances of the GeFS measure for feature selection. The GeFS measure was compared experimentally with two different methods regarding the feature selection capabilities in botnet malware detection: the genetic-algorithm-CFS and the best-first-CFS algorithms. We use five different classifiers to test the detection rates and false positive rates. The experiments show that we can remove 99.9% of irrelevant and redundant features from the datasets, while keeping or yielding even better classification performances. Moreover, the GeFS measure outperforms the genetic-algorithm-CFS and the best-first-CFS methods by removing much more redundant features.

In the field of biometric authentication, automatic signature identification and verification has been a strong research area because of the social and legal acceptance and extensive use of the written signature as an easy method for authentication. Signature verification is a process in which the questioned signature is examined in detail in order to determine whether it belongs to the claimed person or not. Signatures provide a secure means for confirmation and authorization in legal documents. So nowadays, signature identification and verification becomes an essential component in automating the rapid processing of documents containing embedded signatures. Sometimes, part-based signature verification can be useful when a questioned signature has lost its original shape due to inferior scanning quality. In order to address the above-mentioned adverse scenario, we propose a new feature encoding technique. This feature encoding is based on the amalgamation of Gabor filter-based features with SURF features (G-SURF). Features generated from a signature are applied to a Support Vector Machine (SVM) classifier. For experimentation, 1500 (50×30) forgeries and 1200 (50×24) genuine signatures from the GPDS signature database were used. A verification accuracy of 97.05% was obtained from the experiments.

Questioned Document Examination processes often encompass analysis of torn documents. To aid a forensic expert, automatic classification of content type in torn documents might be useful. This helps a forensic expert to sort out similar document fragments from a pile of torn documents. One parameter of similarity could be the script of the text. In this article we propose a method to identify the script in document fragments. Torn documents are normally characterized by text with arbitrary orientation. We use Zernike moment - based feature that is rotation invariant together with Support Vector Machine (SVM) to classify the script type. Subsequently gradient features are used for comparative analysis of results between rotation dependent and rotation invariant feature type. We achieved an overall script-identification accuracy of 81.39% when dealing with 11 different scripts at character/connected-component level and 94.65% at word level.

Feature selection is an important part of a pattern recognition system. A feature selection method is required to be general enough to find representative features from training data, which are then used for classifying test patterns. The situation where the features selected from the training data are quite different from the representative features of the testing data is called over-selecting. The main causes of the over-selecting phenomenon are: non-comprehensive consideration of statistical properties of the training data, heuristic search strategies for feature selection and small sample size of the data set for training. In this paper, we show the influence of the over-selecting phenomenon on the over-fitting phenomenon of machine learning algorithms. We propose a new framework to address principal causes of over-selecting and thus reduce the chance of over-fitting. Our new framework that we call Ensemble Feature Selection measure (EnFS), allows to consider many statistical properties of a given data set at the same time by combining many feature selection methods used in the filter model. From the chosen feature selection measures, a new combined measure is constructed. We also propose a new search algorithm that ensures the globally optimal feature subsets by means of the constructed measure. The new search approach is based on solving a mixed 0–1 linear programming (M01LP) problem by means of the branch-and-bound algorithm. In this M01LP problem, the number of constraints and variables is linear in the number of full set features. In order to evaluate the quality of our EnFS measure, we chose the design of an intrusion detection system (IDS) as a possible application. Experimental results obtained over the KDD CUP'99 benchmarking data set for IDS show that our EnFS measure is capable of reducing over-fitting by addressing over-selecting.

The quality of the feature selection algorithm is one of the most important factors that affects the effectiveness of an intrusion detection system (IDS). Achieving reduction of the number of relevant traffic features without negative effect on classification accuracy is a goal that greatly improves the overall effectiveness of the IDS. Obtaining a good feature set automatically without involving expert knowledge is a complex task. In this paper, we propose an automatic feature selection procedure based on the filter method used in machine learning. In particular, we focus on Correlation Feature Selection (CFS). By transforming the CFS optimization problem into a polynomial mixed 0−1 fractional programming problem and by introducing additional variables in the problem transformed in such a way, we obtain a new mixed 0 − 1 linear programming problem with a number of constraints and variables that is linear in the number of full set features. The mixed 0−1 linear programming problem can then be solved by means of branch-and-bound algorithm. Our feature selection algorithm was compared experimentally with the best-first-CFS and the genetic-algorithm-CFS methods regarding the feature selection capabilities. The classification accuracy obtained after the feature selection by means of the C4.5 and the BayesNet machines over the KDD CUP'99 IDS benchmarking data set was also tested. Experiments show that our proposed method outperforms the best first and genetic algorithm search strategies by removing much more redundant features and still keeping the classification accuracies or even getting better performances.

Arbitrary orientation and sparse data content are common characteristics of torn document. To ensure accuracy and reliability in computer-based analysis, content-zone segmentation is required. In our previous work, we studied segmentation of handwritten and printed text. A questioned document-piece in the form of an office note, however, might also contain non-text data like logos, graphics, and pictures. Hence a more precise content-zone classification is required. In this paper we propose a two-tier approach for non-text, handwriting and printed text segmentation. The first tier aims to discriminate text and non-text regions. The second tier classifies handwritten and printed text within all text zones identified during the first tier. Gabor features and chain-code features are used in Tier-1 and Tier-2, respectively. By using SVM classifier we successfully identified 97.65% of 31,227 text regions in our current test data. The proposed approach identified 98.69% of printed and 96.39% of handwritten text amongst all identified text regions.

Automatic identification of an individual based on his/her handwriting characteristics is an important forensic tool. In a computational forensic scenario, presence of huge amount of text/information in a questioned document cannot be always ensured. Also, compromising in terms of systems reliability under such situation is not desirable. We here propose a system to encounter such adverse situation in the context of Bengali script. Experiments with discrete directional feature and gradient feature are reported here, along with Support Vector Machine (SVM) as classifier. We got promising results of 95.19% writer identification accuracy at first top choice and 99.03% when considering first three top choices.

All Han-based scripts (Chinese, Japanese, and Korean) possess similar visual characteristics. Hence system development for identification of Chinese, Japanese and Korean scripts from a single document page is quite challenging. It is noted that a Han-based document page might also have Roman script in them. A multi-script OCR system dealing with Chinese, Japanese, Korean, and Roman scripts, demands identification of scripts before execution of respective OCR modules. We propose a system to address this problem using directional features along with a Gaussian Kernel-based Support Vector Machine. We got promising results of 98.39% script identification accuracy at character level and 99.85% at block level, when no rejection was considered.

Performance of a pattern recognition system depends strongly on the employed feature-selection method. We perform an in-depth analysis of two main measures used in the filter model: the correlation-feature-selection (CFS) measure and the minimal-redundancy-maximal-relevance (mRMR) measure. We show that these measures can be fused and generalized into a generic feature-selection (GeFS) measure. Further on, we propose a new feature-selection method that ensures globally optimal feature sets. The new approach is based on solving a mixed 0-1 linear programming problem (M01LP) by using the branch-and-bound algorithm. In this M01LP problem, the number of constraints and variables is linear ($O(n)$) in the number $n$ of full set features. In order to evaluate the quality of our GeFS measure, we chose the design of an intrusion detection system (IDS) as a possible application. Experimental results obtained over the KDD Cup'99 test data set for IDS show that the GeFS measure removes 93% of irrelevant and redundant features from the original data set, while keeping or yielding an even better classification accuracy.

Recent results of forgery detection by implementing biometric signature verification methods are promising. At present, forensic signature verification in daily casework is performed through visual examination by trained forensic handwriting experts, without reliance on computer-assisted methods. With this competition on on- and offline skilled forgery detection, our objective is to make a first step towards bridging the gap between automated biometric performances and expert-based visual comparisons. We intent to combine realistic forensic casework with automated methods by testing systems on a forensic-like new dataset. The results achieved by the participating systems are promising: 2.85\% Equal Error Rate (EER) on the online data and 9.15\% on the offline data. From these results we indicate that automated methods might be able to support forensic handwriting experts (FHEs) to formulate the strength of evidence that needs to be reported in court in the future.

A two-stage approach for word-wise identification of English (Roman), Devnagari and Bengali (Bangla) scripts is proposed. This approach balances the tradeoff between recognition accuracy and processing speed. The 1st stage allows identifying scripts with high speed, yet less accuracy when dealing with noisy data. The advanced 2nd stage processes only those samples that yield low recognition confidence in the first stage. For both stages a rough character segmentation is performed and features are computed on segmented character components. Features used in the 1st stage are a 64-dimensional chain-code-histogram feature, while 400-dimensional gradient features are used in the 2nd stage. Final classification of a word to a particular script is done via majority voting of each recognized character component of the word. Extensive experiments with various confidence scores were conducted and reported here. The overall recognition accuracy and speed is remarkable. Correct classification of 98.51% on 11,123 test words is achieved, even when the recognition-confidence is as high as 95% at both stages.

The forensic investigation of a questioned signature written on a piece of paper is a challenging task. Electronic pen-tablets for recording writing movements are considered valuable tools to assist in this effort. However, little is known about the precision and reliability of such electronic devices that are not intended as forensic equipment originally. Moreover, very few studies are conducted on the minimal requirements that would support appropriate forensic signature analysis. This paper presents a systematic study on the accuracy of online pen positions, which extends our previous work on the reliability of pen-force records. Implications of pen-tip displacements are discussed in the context of on- and offline signature data that is produced simultaneously. A basic method and advanced morphing algorithm are detailed that tackle (in)deterministic positioning errors. Experimental results on the processing of 100 authentic signatures and 93 skilled forgeries are reported finally.

In recent years, mathematical, statistical and computer science methods have found extensive application in developing new procedures for crime investigation, prosecution and the enforcement of law. Computer-based methods have also become important tools for performing certain forensic functions. Yet, knowledge and intuition of the human expert plays a central role. This paper gives a brief overview of the potential of computational methods in forensics. The state of the art is described, some practical examples are given and a viewpoint is expressed that the methods can be extended to functions that currently rely only on expert knowledge.

A new search procedure is presented that makes use of the constrained edit distance in the pre-selection of the areas of the digital forensic search space that are interesting for the investigation. The constraints regard the maximum numbers of consecutive deletions and insertions. The original algorithm for computing the constrained edit distance is modified in order to be used in forensic search. The modification removes the constraints before the first substitution and after the last substitution. The advantage of the new method over the use of the unconstrained edit distance in the same procedure lies in the possibility of eliminating the need for finer exhaustive search in the fragments of the search space, in which the detected subsequence is too distorted. It also enables better control over the search process in the case of deliberate distortion of the evidence strings. Experimental results show that the proposed method offers data set reduction in the typical cases of as much as 80% compared to the method that uses unconstrained edit distance.

In this paper, an alternative approach to the non-linear filtering of mathematical morphology for color and multi-spectral images based on the Pareto-dominated hypervolume measure is presented. Parteo-set theory is studied in this context, and among others, successfully implemented in the morphological filtering of color images. The demand to assess the quality of a multi-objective optimization algorithms, in particular, has put forth several kind of measures. One of these measures, the hypervolume, bases on the Lebesque measure of all points dominated by a set of points and maps a set of Pareto-optimal points to a non-negative scalar. By considering the hypervolume in the color-image domain, it can be shown that this hypervolume corresponds to another kind of color morphology, were each pixel in the filtered image represents the hypervolume of its set of neighbours in the original color image. In the following, some properties of the hypervolume, as used as an image-processing filter will be derived, and some potential applications of this approach to color morphology will be suggested.

This paper presents new image processing operations based on the Dubois and Prade (1980) proposal of a fuzzy triangular norm. This definition is recalled, and a computational efficient algorithm is derived for its fast computation. This procedure also gives a comprehensive model for reasoning about the qualitative effects of the Dubois and Prade fuzzy norm on data. The presented image processing operations can be considered as fuzzy morphology operations. Due to this fact the application modes for those operations can be judged. Using the Dubois and Prade fuzzy morphology for background removal on bank cheque images is demonstrated as an application.