Image binarization is fundamental for document image processing. The performance of binarization algorithms depends on several factors that range from the quality of the digitalization devices to the intrinsic features of the document itself and the kind and intensity of the noises present in the image. This assessment on binarizing photographed documents evaluated the quality, time, space, and performance of five new algorithms and ninety-eight "classical" algorithms. The test data set is composed of laser and deskjet printed documents, photographed using six widely used mobile devices with the strobe flash on, off, and in auto modes under three different angles and places of capture.

Many document processing platforms have image binarization as a key step. The performance of binarization algorithms depends on several factors that span from the quality of the digitalization devices to the intrinsic features of the document itself and the kind and intensity of the noises present in the image. Besides that, the processing time is also an important factor that may restrict the applicability of a given algorithm. This competition on binarizing photographed documents assessed the quality, time, space, and performance of twenty three new algorithms and seventy-five "classical" algorithms. The test dataset is composed of offset and deskjet printed documents, photographed using six widely-used mobile devices with the strobe flash on, off, and in auto modes under two different angles and places of capture.

Automatic Text Summarization (ATS) is a research area that originated in the late 1950s and has gained increasing importance with the surging amount of text data available today. One of the key challenges in this area is how to quantitatively assess the quality of the summaries produced. The three most widely quantitative measures used for this task are: ROUGE, BLEU and BERTScore. This paper attempts to comparatively evaluate the validity and reliability of such measures. The concept of Shannon' entropy from information theory served as background for this work. Experiments were conducted using the CNN corpus, focusing on news articles written in English.

Love it or hate it, the peer review process (whether open, blind, or even double-blind) has become the standard and accepted way of assessing the quality of papers before publication, be it for a conference, journal, or book. Indeed, forming the program committee is an essential part in any conference organisation and a good program committee may well be the differentiator from peer conferences. However, we have all been the recipients of a less than stellar/helpful review: from the snarky ones to the one-liners, these reviews can be demoralising and can give the peer-review process a bad reputation! The scope of this tutorial is then to encourage researchers to become more involved in the peer-review process by joining program committees and encourages good practices to collectively strengthen the quality of the peer-review process.

Document image binarization is a fundamental step in many document processes. No binarization algorithm performs well on all types of document images, as the different kinds of digitalization devices and the physical noises present in the document and acquired in the digitalization process alter their performance. Besides that, the processing time is also an important factor that may restrict its applicability. This competition on binarizing photographed documents assessed the quality, time, space, and performance of five new algorithms and sixty-four "classical" and alternative algorithms. The evaluation dataset is composed of laser and deskjet printed documents, photographed using six widely-used mobile devices with the strobe flash on and off, under two different angles and places of capture.

While much work on abstractive summarization has been conducted in recent years, including state-of-the-art summarizations from GPT-4, extractive summarization's lossless nature continues to provide advantages, preserving the style and often key phrases of the original text as meant by the author. Libraries for extractive summarization abound, with a wide range of efficacy. Some do not perform much better or perform even worse than random sampling of sentences extracted from the original text. This study breathes new life to using classical algorithms by proposing parallelism through an implementation of a second order meta-algorithm in the form of the Tessellation and Recombination with Expert Decisioner (T&R) pattern, taking advantage of the abundance of already-existing algorithms and dissociating their individual performance from the implementer's biases. Resulting summaries obtained using T&R are better than any of the component algorithms.

Determining the relationships between characters is an important step in analyzing fictional works. Knowing character relationships can be useful when summarizing a work and may also help to determine authorship. In this paper, scores are generated for pairs of characters in fictional works, which can be used for classification tasks if characters have a relationship or not. An SVM is used to predict relationships between characters. Characters farther from the decision boundary often had stronger relationships than those closer to the boundary. The relative rank of the relationships may have additional literary and authorship related purposes.

Our vision is to achieve societally responsible secure and trustworthy cyberspace that puts algorithmic and technological checks and balances on the indiscriminate sharing and analysis of data. We achieve this vision in a holistic manner by framing research directions with four major considerations: (i) Expanding knowledge and understanding of security and privacy perceptions and expectations in vulnerable groups, which significantly contribute to their unwillingness to share data, and use that knowledge to drive research in (a) mitigating missing/imbalanced data problems, (b) understanding and modeling security and privacy risks of data sharing, and (c) modeling utility of data sharing. (ii) Developing a risk-adaptive, policy model capable of capturing and articulating security and privacy expectations of users that are relevant in a particular context and develops associated technology to ensure provenance and accountability. (iii) Developing robust AI/ML algorithms that are transparent and explainable with respect to fairness and bias to reduce/eliminate discrimination, misuse, privacy violations, or other cyber-crimes. (iv) Developing models and techniques for a nuanced, contextually adaptive, and graded privacy paradigm that allows trade-offs between privacy and utility. Towards this, in this paper we present the SAFE-PASS framework to provide Stewardship, Advocacy, Fairness and Empowerment in Privacy, Accountability, Security, and Safety for Vulnerable Groups.

Today, over eighty percent of the world's population owns a smart-phone with an in-built camera, and they are very often used to photograph documents. Document binarization is a key process in many document processing platforms. This competition on binarizing photographed documents assessed the quality, time, space, and performance of five new algorithms and sixty-four "classical" and alternative algorithms. The evaluation dataset is composed of offset, laser, and deskjet printed documents, photographed using six widely-used mobile devices with the strobe flash on and off, under two different angles and places of capture.

Chemical Safety Data Sheets (SDS) are the primary method by which chemical manufacturers communicate the ingredients and hazards of their products to the public. These SDSs are used for a wide variety of purposes ranging from environmental calculations to occupational health assessments to emergency response measures. Although a few companies have provided direct digital data transfer platforms using xml or equivalent schemata, the vast majority of chemical ingredient and hazard communication to product users still occurs through the use of millions of PDF documents that are largely loaded through manual data entry into downstream user databases. This research focuses on the reverse engineering of SDS document types to adapt to various layouts and the harnessing of meta-algorithmic and neural network approaches to provide a means of moving industrial institutions towards a digital universal SDS processing methodology. The complexities of SDS documents including the lack of format standardization, text and image combinations, and multi-lingual translation needs, combined, limit the accuracy and precision of optical character recognition tools.

Smartphones with cameras are omnipresent in today's world and are very often used to photograph documents. Document binarization is a key process in many document processing platforms. This competition on binarizing photographed documents assessed the quality and time performance of 13 new algorithms and 50 existing algorithms. The evaluation dataset is composed of offset, laser, and deskjet printed documents, photographed using four widely-used mobile devices with the strobe flash on and off, under two different angles and places of capture.

We present an overview of the field of malware analysis with emphasis on issues related to document engineering. We will introduce the field with a discussion of the types of malware, including executable binaries, malicious PDFs, polymorphic malware, ransomware, and exploit kits. We will conclude with our view of important research questions in the field. This is an updated version of tutorials presented in previous years, with more information about newly-available tools.

The DocEng'2020 Competition on Extractive Text Summarization assessed the performance of six new methods and fourteen classical algorithms for extractive text sumarization. The systems were evaluated using the CNN-Corpus, the largest test set available today for single document extractive summarization using two different strategies and the ROUGE and the direct match measures.

Document image binarization is a key process in many document processing platforms. The DocEng'2020 Time-Quality Competition on Binarizing Photographed Documents assessed the performance of eight new algorithms and also 41 other "classical" algorithms. Besides the quality of the binary image, the execution time of the algorithms was assessed. The evaluation dataset is composed of 32 documents photographed using four widely-used mobile devices with the strobe flash on and off, under several different angles of capture.

The DocEng'19 Competition on Extractive Text Summarization assessed the performance of two new and fourteen previously published extractive text sumarization methods. The competitors were evaluated using the CNN-Corpus, the largest test set available today for single document extractive summarization.

This paper details the features and the methodology adopted in the construction of the CNN-corpus, a test corpus for single document extractive text summarization of news articles. The current version of the CNN-corpus encompasses 3,000 texts in English, and each of them has an abstractive and an extractive summary. The corpus allows quantitative and qualitative assessments of extractive summarization strategies.

This paper details the development and features of the CNN-corpus in Spanish, possibly the largest test corpus for single document extractive text summarization in the Spanish language. Its current version encompasses 1,117 well-written texts in Spanish, each of them has an abstractive and an extractive summary. The development methodology adopted allows good-quality qualitative and quantitative assessments of summarization strategies for tools developed in the Spanish language.

We present some results from a joint project between HP Labs, Cardiff University and Dyfed Powys Police on predictive policing. Applications of the various techniques from recommender systems and text mining to the problem of crime patterns recognition are demonstrated. Our main idea is to consider crime records for different regions and time period as a corpus of text documents with words being crime types. We apply tools from NLP and text documents classifications to analyse different regions in time and space. We evaluate performance of several measures of similarity for texts and documents clustering algorithms.

In this tutorial, we consider important aspects (algorithms, approaches, considerations) for tagging both unstructured and structured text for downstream use. This includes summarization, in which text information is compressed for more efficient archiving, searching, and clustering. In the tutorial, we focus on the topic of automatic text summarization, covering the most important milestones of the six decades of research in this area.

Document engineering is all about building systems and tools that allow people to work with documents and document collections. A key aspect is the usefulness and usability of these tools. In this tutorial, we will look at the many different kinds of user studies and user evaluations that can be used to inform the design and improve utility and usability of document engineering applications. The tutorial will be based on actual studies and will also give participants a chance to explore how they might use these techniques in their research or system development.

In the world of ground truthing--that is, the collection of highly valuable labeled training and validation data-there is a tendency to follow the path of first training on a set of data, then validating the data, and then testing the data. However, in many cases the labeled training data is of non-uniform quality, and thus of non-uniform value for assessing the accuracy and other performance indicators for analytics algorithms, systems and processes. This means that one or more of the so-labeled classes is likely a mixture of two or more clusters or sub-classes. These data may inhibit our ability to assess the classifier to use for deployment. We argue that one must learn about the labeled data before the labeled data can be used for downstream machine learning; that is, we reverse the validation and training steps in building the classifier. This "learning before learning" is assessed using a CNN corpus (cnn.com) which was hand-labeled as comprising 12 classes. We show how the suspect classes are identified using the initial validation, and how training after validation occurs. We then apply this process to the CNN corpus and show that it consists of 9 high-quality classes and three mixed-quality classes. The effects of this validation-training approach is then shown and discussed.

For distance learning applications, inferring the cognitive states of students, particularly, their concentration and comprehension levels during instruction, is important to assess their learning efficacy. In this paper, we investigated the feasibility of using EEG recordings generated from an off-the-shelf, wearable device to automatically classify the cognitive states of students as they were asked to perform a series of reading and question answering tasks. We showed that the EEG data can effectively predict whether a student is attentive or distracted as well as the student's reading speed, which is an important measure of reading fluency. However, the EEG signals alone are insufficient to predict how well the students can correctly answer questions related to the reading materials as there were other confounding factors, such as the students' background knowledge, that must be taken into consideration. We also showed that the accuracy in predicting the different cognitive states depends on the choice of classifier used (global, local, or multi-task learning). For example, the concentration level of a student can be accurately predicted using a local model whereas a global model that incorporates side information about the student's background knowledge is more effective at predicting whether the student will correctly answer questions about the materials they read.

This paper presents a new method for improving the cohesiveness of summaries generated by extractive summarization systems. The solution presented attempts to improve the legibility and cohesion of the generated summaries through coreference resolution. It is based on a post-processing step that binds dangling coreference to the most important entity in a given coreference chain. The proposed solution was evaluated on the CNN corpus of 3,000 news articles, using four state-of-the-art summarization systems and seventeen techniques for sentence scoring proposed in the literature. The experimental results may be considered encouraging, as the final summaries reached better ROUGE scores, besides being more cohesive.

Mobile devices such as smart phones and tablets are omnipresent in modern societies. Such devices allow browsing the Internet. This paper briefly describes two tools for news article summarization in mobile devices that attempts to automatically collect and sieve the most important information of news article in WebPages.

The existing automatic text summarization systems whenever applied to web-pages of news articles show poor performance as the text is encapsulated within a HTML page. This paper takes advantage of the link identification and content extraction techniques. The results show the validity of such a strategy.

Today, students are offered a wide variety of alternatives to printed material for the consumption of educational content. Previous research suggests that, while digital content has its advantages, printed content still offers benefits that cannot be matched by digital media. This paper introduces the Meaningful Education and Training Information System (METIS), a multi-faceted hybrid book learning platform. The goal of the system is to provide an easy digital-to-print-to-digital content creation and reading service. METIS incorporates technology for layout, personalization, co-creation and assessment. These facilitate and, in many cases, significantly simplify common teacher/student tasks. Our system has been demonstrated at several international education events, partner engagements, and pilots with local universities and high schools. We present the system and discuss how it enables hybrid learning.

Some of the recent state-of-the-art systems for Automatic Text Summarization rely on the concept-based approach using Integer Linear Programming (ILP), mainly for multi-document summarization. A study on the suitability of such an approach to single-document summarization is still missing, however. This work presents an assessment of several methods of concept weighing for a concept-based ILP approach on the single-document summarization scenario. The unigram and bigram representations for concepts are also investigated. The experimental results obtained on the DUC 2001-2002 and the CNN corpora show that bigrams are more suitable than unigrams for the representation of concepts. Among the concept scoring methods investigated, the sentence position method presented the best performance on all evaluation corpora.

Analytics obtained during the creation of a database of mass serialized codes can also be used to help enforcement of encryption policy on documents. In this paper, we introduce a set of metrics which complement traditional NIST cryptography methods -- 4 mass serialization and one entropy metric -- which in combination can allow a discrimination between encrypted vs. zipped files. We describe the use of these methods to identify a broad range of non-randomness in number sets, and apply them to a more mundane problem-that of automatic assessment of the encryption state of a corpora of documents.

Text summarization is the process of automatically creating a shorter version of one or more text documents. This paper presents a qualitative and quantitative assessment of the 22 state-of-the-art extractive summarization systems using the CNN corpus, a dataset of 3,000 news articles.

An efficient way to automatically classify documents may be provided by automatic text summarization, the task of creating a shorter text from one or several documents. This paper presents an assessment of the 15 most widely used methods for automatic text summarization from the text classification perspective. A naive Bayes classifier was used showing that some of the methods tested are better suited for such a task.

The need for automatic generation of summaries gained importance with the unprecedented volume of information available in the Internet. Automatic systems based on extractive summarization techniques select the most significant sentences of one or more texts to generate a summary. This article makes use of Machine Learning techniques to assess the quality of the twenty most referenced strategies used in extractive summarization, integrating them in a tool. Quantitative and qualitative aspects were considered in such assessment demonstrating the validity of the proposed scheme. The experiments were performed on the CNN-corpus, possibly the largest and most suitable test corpus today for benchmarking extractive summarization strategies.

Given a document collection, existing systems allow users to browse the collection or perform searches that return lists of documents ranked based on their relevance to the user query. While these approaches work fine when a user is trying to locate specific documents, they are insufficient when users need to access the pertinent documents in some logical order, for example for learning or editorial purposes. We present a system that automatically organizes a collection of documents in a tree from general to more specific documents, and allows a user to choose a reading sequence over the documents. This a novel way to content consumption that departs from the typical ranked lists of documents based on their relevance to a user query and from static navigational interfaces. We present a set of algorithms that solve the problem and we evaluate their performance as well as the reading trees generated.

Text summarization is the process of creating a shorter version of one or more text documents. Automatic text summarization has become an important way of finding relevant information in large text libraries or in the Internet. Extractive text summarization techniques select entire sentences from documents according to some criteria to form a summary. Sentence scoring is the technique most used for extractive text summarization, today. Depending on the context, however, some techniques may yield better results than some others. This paper advocates the thesis that the quality of the summary obtained with combinations of sentence scoring methods depend on text subject. Such hypothesis is evaluated using three different contexts: news, blogs and articles. The results obtained show the validity of the hypothesis formulated and point at which techniques are more effective in each of those contexts studied.

The text data available on the Internet is not only huge in volume, but also in diversity of subject, quality and idiom. Such factors make it infeasible to efficiently scavenge useful information from it. Automatic text summarization is a possible solution for efficiently addressing such a problem, because it aims to sieve the relevant information in documents by creating shorter versions of the text. However, most of the techniques and tools available for automatic text summarization are designed only for the English language, which is a severe restriction. There are multilingual platforms that support, at most, 2 languages. This paper proposes a language independent summarization platform that provides corpus acquisition, language classification, translation and text summarization for 25 different languages.

Sentence similarity is used to measure the degree of likelihood between sentences. It is used in many natural language applications, such as text summarization, information retrieval, text categorization, and machine translation. The current methods for assessing sentence similarity represent sentences as vectors of bag of words or the syntactic information of the words in the sentence. The degree of likelihood between phrases is calculated by composing the similarity between the words in the sentences. Two important concerns in the area, the meaning problem and the word order, are not handled, however. This paper proposes a new sentence similarity assessment measure that largely improves and refines a recently published method that takes into account the lexical, syntactic and semantic components of sentences. The new method proposed here was benchmarked using a publically available standard dataset. The results obtained show that the new similarity assessment measure proposed outperforms the state of the art systems and achieve results comparable to the evaluation made by humans.

Relation extraction (RE) aims at finding the way entities, such as person, location, organization, date, etc., depend upon each other in a text document. Ontology Population, Automatic Summarization, and Question Answering are fields in which relation extraction offers valuable solutions. A relation extraction method based on inductive logic programming that induces extraction rules suitable to identify semantic relations between entities was proposed by the authors in a previous work. This paper proposes a method to simplify graph-based representations of sentences that replaces dependency graphs of sentences by simpler ones, keeping the target entities in it. The goal is to speed up the learning phase in a RE framework, by applying several rules for graph simplification that constrain the hypothesis space for generating extraction rules. Moreover, the direct impact on the extraction performance results is also investigated. The proposed techniques outperformed some other state-of-the-art systems when assessed on two standard datasets for relation extraction in the biomedical domain.

A general two-dimensional coding method is presented that allows recovery of data based on only a cropped portion of the code, and without knowledge of the carrier image. A description of both an encoding and recovery system is provided. Our solution involves repeating a payload with a fixed number of bits, assigning one bit to every symbol in the image - whether that symbol is data carrying or non-data carrying - with the goal of guaranteeing recovery of all the bits in the payload. Because the technique is applied to images, for aesthetic reasons we do not use fiducials, and do not employ any end-of-payload symbols. The beginning of the payload is determined by a phase code that is interleaved between groups of payload rows. The recovery system finds the phase row by evaluating candidate rows, and ranks confidence based on the sample variance. The target application is data-bearing clustered-dot halftones, so special consideration is given to the resulting checkerboard subsampling. This particular application is examined via exhaustive simulations to quantify the likelihood of unrecoverable bits and bit redundancy as a function of offset, crop window size, and phase code spacing.

Text summarization is the process of automatically creating a shorter version of one or more text documents. In this context, word-based, sentence-based and graph-based methods approaches are largely used. Among these, graph based methods for automatic text summarization produce summaries based on the relationships between sentences. These relationships may also support the creation of several text processing applications such as extractive and abstractive summaries, question-answering and information retrieval systems, among others. A new graph model for text processing applications is proposed in this paper. It relies on four dimensions (similarity, semantic similarity, co reference, discourse information) to create the graph. The rationale behind the proposal presented here is resorting to more dimensions than previous works, and taking into account co reference resolution, taking into account to the role of pronouns in connecting the sentences. Co reference was not used in any previous graph based summarization technique. An experiment was performed using the Text Rank algorithm with the presented approach, on the CNN corpus. The results show that the model proposed here outperforms the current approaches both quantitatively and qualitatively.

Electrocardiograms (ECGs) are prone to significant noise, including breathing and motion artifacts. Generally, for redundant lead systems, the noise is often correlated between leads. Using this redundancy, a simple image analysis pipeline was devised to improve the appearance of ECGs: (1) convert the ECG into a vectorcardiogram (VCG) representation, (2) filter in the VCG domain using a simple moving average filter, and (3) convert the VCG back into an ECG. The resulting ECG shows significantly improved P and T wave quality without loss of high frequency information in the QRS wave of the ECG. The implications of this approach for automated ECG monitoring and interpretation are described.

At the microscopic level, printing on a substrate exhibits imperfections that can be used as a unique identifier for labels, documents and other printed items. In previous work, we have demonstrated using these minute imperfections around a simple forensic mark such as a single printed character for robust authentication of the character with a low cost (and mobile) system. This approach allows for product authentication even when there is only minimal printing (e.g. on a small label or medallion), supporting a variety of secure document workflows. In this paper, we present an investigation on the influence that the substrate type has on the imperfections of the printing process that are used to derive the character 'signature'. We also make a comparison between two printing processes, dry electro photographic process (laser) and (thermal) inkjet. Understanding the sensitivity of our methods to these factors is important so that we know the limitations of the approach for document forensics.

Automatic text summarization is an important and challenging problem. Over the years, the amount of text available electronically has grown exponentially. This growth has created a huge demand for automatic methods and tools for text summarization. We can think of automatic summarization as a type of information compression. To achieve such compression, better modelling and understanding of document structures and internal relations is required. In this article, we develop a novel approach to extractive text summarization by modelling texts and documents as small-world networks. Based on our recent work on the detection of unusual behavior in text, we model a document as a one-parameter family of graphs with its sentences or paragraphs defining the vertex set and with edges defined by Helmholtz's principle. We demonstrate that for some range of the parameters, the resulting graph becomes a small-world network. Such a remarkable structure opens the possibility of applying many measures and tools from social network theory to the problem of extracting the most important sentences and structures from text documents. We hope that documents will be also a new and rich source of examples of complex networks.

Recently-introduced Publicly Posted Composite Documents (PPCDs) enable composite documents with different formats and differential access control to participate in cross-organizational workflows distributed over potentially non-secure channels. The original PPCD design was based on a Public Key Infrastructure, requiring each workflow participant to own a pair of public and private keys. This solution also required the document master to know the corresponding valid public keys (certificates) of all participants prior to commencement of the workflow. Using Identity Based Encryption (IBE), a recently described cryptographic technique, we eliminate the requirement for the prior knowledge and distribution of the workflow participants' keys. The required public keys for each workflow participant are calculated based on user identities and other relevant factors at workflow onset. The generation of corresponding private keys can be delayed up until the workflow step, when the corresponding workflow participants require access to the document. The solution presented provides automatic workflow order enforcement and the ability to impose multiple document release dates real-time.

With the boom in interactive and composite documents and the increased coupling between the on-line and physical worlds, the need for secure document engineering is greater than ever. Four important factors contribute to the need to re-engineer document lifecycles and the associated workflows. The first is the rapid increase in mobile access to documents. The second is the movement of documents from private directories, shared directories and intranets to the cloud. The third is the increased generation of - and expectation for - document content, context and use analytics. Finally, the proliferation of social website applications and services over the past half-decade have created for many a constant state of log-in. Each of these trends creates a significantly increased "attack surface" for individuals, organizations and governments interested in breaching the privacy and security of web users. Combined, these transformations create as big a change to content security as that of the browser in the 1990s. In this workshop, we consider the impact of these ongoing transformations in document creation and interaction, and consider what the best approaches will be to provide privacy and security in light of these transformations.

In this paper, we describe the feasibility and practical study of the recently proposed idea for data leak prevention (DLP) based on end-point policy enforcement. The most reassuring way to prevent sensitive data leak is to thwart sensitive data export before it has a chance to occur. Using a System Call Interception (SCI) technique we investigate the possibility of automatically detecting and amending a non-desired, policy breaching behavior at the "intention" stage: as the corresponding system call is called by an application, but before the action has been accomplished. The SCI method is especially valuable for "black box" applications, for which source code is not available. In our system, we catalog the system calls involved in the DLP events, and reduce our SCI to the minimum necessary set of system calls associated with the sensitive, DLP-requiring tasks. We describe the system behavior for several different applications that we have studied to date.

True digital photos and the digital images of scanned photographs have very different properties. The illumination pattern and palette of the two kinds of images are different. Being able to distinguish between them is important, as each of these should be handled during printing with a class-specific pipeline of image transformation algorithms, and misclassification results in detrimental imaging effects. This paper presents an automatic classifier to discriminate between the two sources. The classifier proposed is fast enough to be embedded in the driver of any printing device today.

In this paper we describe the possibility of constructing the well-known small world topology for an ordinary document, based on the actual document structure. Sentences in such a graph are represented by nodes, which are connected if and only if the corresponding sentences are neighbors or share at least one common keyword. This graph is built using a carefully selected one-parameter set of keywords. By varying this parameter - the level of meaningfulness - we transition the document-representing graph from a trivial path graph into a large random graph. During such a conversion, as the parameter is varied over its range, the graph becomes a small world. This in turn opens the possibility of applying many well-established ranking algorithms to the problem of ranking sentences and paragraphs in text documents. These rankings are, in turn, crucial for document understanding, summarization and information extraction. These graphs can also serve as a source of interesting small world graphs for the theory of complex networks.

Publicly Posted Composite Documents (PPCDs) address the problem of composite documents with different formats and varied sensitivity participating in cross-organizational workflows distributed over potentially non-secure channels. An early version of the PPCD, however, is susceptible to silent workflow termination, wherein a document shipped by one workflow participant is never received by the subsequent participant, causing the workflow to terminate without notification. In the current paper, we extend the PPCD solution to provide an early detection mechanism for a potential workflow termination, to provide workflow recovery procedures, and to provide for the often mandatory document logging and audit information.

Recently-introduced Publicly Posted Composite Documents (PPCDs) address the problem of composite documents with different formats and differential access control participating in cross-organizational workflows distributed over potentially non-secure channels. An early version of the PPCD only considered two simplest workflow types: ordered and unordered. In real life, however, fragments of such pure types are likely to be combined into mixed workflows with interleaved ordered and unordered workflow steps. In the current paper, we introduce a payload matrix for a generic mixed workflow. We provide a computationally beneficial solution for enforcing access in ordered workflows by reducing the volume of data that needs to be encrypted and then subsequently decrypted. Furthermore, we address the problem of enforcing order in transitions between different types of workflow steps: from an unordered step to ordered steps and vice versa, and from one unordered step to another. This prevents the participants of any workflow step from being able to access a PPCD document prior to their workflow step.

Recent advancements in document handling tools have greatly simplified the tasks of reusing content and copying between secure and unsecure locations. As a result, intentional and unintentional data leaks are occurring more often and so posing a serious risk. To address this issue, context-aware policies have been introduced so that the policies pertain to the patterns of text contained in actual run-time content - as well as static document-level metadata with the aim of distinguishing different types of sensitive documents. Unlike traditional metadata based policies, context-aware policies do not need to have any overlap in their conditions for contradictions to occur, which makes it harder to create and manage coherent, ambiguity free sets of policies. We provide a novel Intelligent Modeling Assistant (IMA) that automatically deals with resolvable complications in modelling, while breaking tasks requiring manual intervention into a set of simple decisions. The IMA also helps the administrator to understand possible consequences of policies through sets of examples. Our implementation using constraint satisfaction technology is described.

Keyword extraction is a fundamental problem in text data mining and document processing. A large number of document processing applications directly depend on the quality and speed of keyword extraction algorithms. In this article, a novel approach to rapid change detection in data stream.

A novel mechanism for providing and enforcing differential access control for publicly-posted composite documents is proposed. The concept of a document is rapidly changing: individual file-based, traditional formats can no longer accommodate the required mixture of differently formatted parts: individual images, video/audio clips, PowerPoint presentations, html-pages, Word documents, Excel spreadsheets, pdf files, etc. Multi-part composite documents are created and managed in complex workflows, with participants including external consultants, partners and customers distributed across the globe, with many no longer contained within one monolithic secure environment. Distributed over non-secure channels, these documents carry different types of sensitive information: examples include (a) an enterprise pricing strategy for new products, (b) employees' personal records, (c) government intelligence, and (d) individual medical records. A central server solution is often hard or impossible to create and maintain for ad-hoc workflows. Thus, the documents are often circulated between workflow participants over traditional, low security e-mails, placed on shared drives, or exchanged using CD/DVD or USB. The situation is more complicated when multiple workflow participants need to contribute to various parts of such a document with different access levels: for example, full editing rights, read-only, reading of some parts only, etc., for different users. We propose a full scale differential access control approach, enabling public posting of composite documents, to address these concerns.

The changing nature of document workflows, document privacy and document security merit a new approach to the enforcement of policy. We propose the use of automated means for enforcing policy, which provides advantages for compliance and auditing, adaptability to changes in policy, and compatibility with a cloud-based exchange. We describe the Automated Policy Enforcement eXchange (APEX) software system, which consists of: (1) a policy editor, (2) a policy server, (3) a local daemon on every PC/laptop to maintain local secure up-to-date storage and policy, and (4) local (policy-enforcing) wrappers to capture document-handling user actions such as document export, e-mail, print, edit and save. During the performance of relevant incremental change, or other user-elicited action, on a composite document, the document and its metadata are scanned for salient policy eliciting terms (PETs). The document is then partitioned based on relevant policies and the security policy for each part is determined. If the document contains no PETs, then the user-initiated actions are allowed; otherwise, alternative actions are suggested, including: (a) encryption, (b) redirecting to a secure printer and requiring authorization (e.g. PIN) for printing, and (c) disallowing printing until specific sensitive data is removed.

Much of the focus in document security tends to be on the deterrent -- the physical (printed, manufactured) item placed on a document, often used for routing in addition to security purposes. Hybrid (multiple) deterrents are not always reliably read by a single imaging device, and so a single device generally cannot simultaneously provide overall document security. We herein show how a relatively simple deterrent can be used in combination with multiple imaging devices to provide document security. In this paper, we show how these devices can be used to classify the printing technology used, a subject of importance for counterfeiter identification as well as printer quality control. Forensic-level imaging is also useful in preventing repudiation and forging, while mobile and/or simple scanning can be used to prevent tampering -- propitiously in addition to providing useful, non-security related, capabilities such as document routing (track and trace) and workflow association.

This paper describes the use of a 1:1 magnification 3.5 micron true resolution Dyson Relay lens-based 3 MPixel USB CMOS imaging device (DR CID) and software forensic image analysis system. The device enables the simultaneous capture of both intentional printing shapes and unintentional printing artifacts caused by the printing process and the interaction of the ink with the substrate on which printing occurs. The custom image analysis system written for the DR CID device allows even a single printed character to simultaneously provide fiducial marking, inspection information, authentication and forensics. We report herein on the sensitivity of the system and initial results for the reliable authentication of a printed character using DR CID hardware devices.

In high-speed printed media inspection environments, image restoration pipelines play a critical role in establishing and continually evaluating performance. A key role of such systems is to understand, mitigate (and possibly remove) artifacts introduced by motion blur. An approach is proposed that uses barcodes to help calibrate a one-dimensional blur restoration pipeline. Techniques are demonstrated whereby the structure of barcode markings may be leveraged to estimate motion blur parameters, even under extreme blur conditions or when the barcode is unknown. In addition, a framework for comparing blur estimation procedures based on barcode readability is introduced. These techniques can be applied independently of one another, but together form a set of useful tools for blur restoration pipeline calibration. Within this framework, it is shown that a low-complexity blur estimation strategy demonstrates performance competitive with state-of-the-art approaches in term of speed and accuracy.

Forensic analysis of individual printed items, including single characters, provides a readily-integrated means to extend security to any printed item (label, document, package, etc.). In this paper we demonstrate, for the first time, forensic levels of image inspection for workflows involving two very different high-resolution imaging devices, the Dyson Relay CMOS Imaging Device (DrCID) and a high speed line-scan camera. In particular, we show how a similarity metric can be used to identify specific characters with less than 1 in 109chance of false matching, while closing the loop between in-line (production) using the line-scan camera and end-user (investigative) inspection with DrCID.

2D barcodes are taking on increasing significance as the ubiquity of high-resolution cameras, combined with the availability of variable data printing, drives increasing amounts of click and connect applications. Barcodes therefore serve as an increasingly significant connection between physical and electronic portions, or versions, of documents. The use of color provides many additional advantages, including increased payload density and security. In this paper, we consider four factors affecting the readable payload in a color barcode: (1) number of print-scan (PS), or copy, cycles, (2) image restoration to offset PS-induced degradation, (3) the authentication algorithm used, and (4) the use of spectral pre-compensation (SPC) to optimize the color settings for the color barcodes. The PS cycle was shown to consistently reduce payload density by approximately 55% under all tested conditions. SPC nearly doubled the payload density, and selecting the better authentication algorithm increased payload density by roughly 50% in the mean. Restoration, however, was found to increase payload density less substantially (~30%), and only when combined with the optimized settings for SPC. These results are also discussed in light of optimizing payload density for the generation of document security deterrents.

Functional image classification is the assignment of different image types to separate classes to optimize their rendering for reading or other specific end task, and is an important area of research in the publishing and multi-Average industries. This paper presents recent research on optimizing the simultaneous classification of documents, photos and logos. Each of these is handled during printing with a class-specific pipeline of image transformation algorithms, and misclassification results in pejorative imaging effects. This paper reports on replacing an existing classifier with a Weka-based classifier that simultaneously improves accuracy (from 85.3% to 90.8%) and performance (from 1458 msec to 418 msec/image). Generic subsampling of the images further improved the performance (to 199 msec/image) with only a modest impact on accuracy (to 90.4%). A staggered subsampling approach, finally, improved both accuracy (to 96.4%) and performance (to 147 msec/image) for the Weka-base classifier. This approach did not appreciable benefit the HP classifier (85.4% accuracy, 497 msec/image). These data indicate staggered subsampling using the optimized Weka classifier substantially improves the classification accuracy and performance without resulting in additional “egregious” misclassifications (assigning photos or logos to the “document” class).

Document authentication is the process by which a unique identifier is assigned to each version of a printed document and is verified via an inspection procedure. One type of approach that has proven successful involves printing overt color-tile security deterrents, which can be scanned and analyzed to provide authenticity. For a variety of reasons discussed herein, it is advantageous to estimate whether or not a printed deterrent will authenticate without invoking the actual authentication process. Several algorithms are presented to predict the outcome of the authentication process. The area-over-the-curve (AOC) statistic is one of the tools used to characterize how well each candidate algorithm estimates authentication performance in the presence of distortions introduced via different print+scan paths. A surprising but useful result is that a no-reference metric yields the best performance: the AOC is approximately half that of all other tested methods, and is over five times smaller than the AOC achieved by state-of-the-art algorithms SSIM and VIF.

The process of adding high-density information onto printed material enables and improves interesting hardcopy document applications, such as: security, authentication, physical-electronic round tripping, item-level tagging as well as consumer/product interaction. This investigation on robust and high capacity print codes aims to maximize information payload in a given printed page area, subject to robustness to distortions originated by printing and scanning processes and also to degradations introduced by user manipulation of printed documents. The novel approach includes statistical print-and-scan channel characterization, designing of robust segmentation, unsupervised Bayesian color classification with expectation-maximization algorithm for parameters estimation of a mixture of Gaussians model and design of error correction codes. Results illustrate the performance evaluated under real channel and distortions conditions. High payload is achieved with sufficient robustness to distortions resulting of regular office hardcopy document handling: print-and-scan channel and user manipulation.

Adding high-density information to printed materials enables and improves interesting hardcopy document applications involving security, authentication, physical-electronic round tripping, item-level tagging, and consumer/product interaction. This investigation of robust and high capacity print codes aims to maximize information payload in a given printed page area, subject to robustness to channel errors including distortions introduced by the printing and scanning processes and also due to the usual degradations introduced by user manipulation of printed documents. The novel approach includes statistical print-and-scan channel characterization, designing of robust segmentation using visual cues, unsupervised Bayesian color classification with expectation-maximization algorithm for parameters estimation of a mixture of Gaussians model and design of error correction codes. Results illustrate the performance evaluated under real channel and distortions conditions. High payload is achieved with sufficient robustness to distortions resulting of regular office hardcopy document handling: print-and-scan channel and user manipulation.

Atmospheric turbulence is caused by the random fluctuations of the refraction index of the medium. It can lead to blurring in images acquired from a long distance away. Since the degradation is often not completely known, the problem is viewed as blind image deconvolution or blur identification. Our previous work has observed that blurring increases kurtosis and introduced a new blur identification method based on kurtosis minimization (KM). In this letter, this observation has been studied using phase correlation. The KM method is compared with two other signal processing methods. The limitation of the method is also discussed.

Pre-processing for raster image based document segmentation begins with image thresholding, which is a binarization process separating foreground from background. In this paper, we compare an existing (Otsu), modified existing (Kittler-Illingworth) and simple peak-based thresholding approach on a set of 982 documents for which existing ground truth (full text) is available. We use the output of an open source OCR engine which incorporates an adaptive/dynamic thresholder that can be bypassed by one of the three global thresholds we tested. This allowed comparison of these three approaches in the aggregate. We then used an independently-generated dictionary as a means of characterizing thresholder efficacy. Such an approach, if successful, will provide the means for selecting an optimal thresholder in the absence of a large set of ground truthed documents. Our preliminary findings here indicate that this approach may provide a reliable means for thresholder comparison and eventually preclude the need for time-intensive human ground truthing.

In this paper, an example-based image denoising algorithm is introduced. Image denoising is formulated as a regression problem, which is then solved using support vector regression (SVR). Using noisy images as training sets, SVR models are developed. The models can then be used to denoise different images corrupted by random noise at different levels. Initial experiments show that SVR can achieve a higher peak signal-to-noise ratio (PSNR) than the multiple wavelet domain Besov ball projection method on document images.

Motivated by the success of support vector regression (SVR) in blind image deconvolution, we apply SVR to single-frame super-resolution. Initial results show that even when trained on as little as a single image, SVR is able to learn a generally applicable model that can super-resolve dissimilar images.

Our earlier work revealed a connection between blind image deconvolution and principal components analysis (PCA). In this letter, we explicitly formulate multichannel and single-channel blind image deconvolution as a PCA problem. Although PCA is derived from blur models that do not contain additive noise, it can be justified on both theoretical and experimental grounds that the PCA-based restoration algorithm is actually robust to the presence of white noise. The algorithm is applied to the restoration of atmospheric turbulence-degraded imagery and compared to an adaptive Lucy-Richardson maximum-likelihood algorithm on both real and simulated atmospheric turbulence blurred images. It is shown that the PCA-based blind image deconvolution runs faster and is more robust to noise.

This letter introduces a new algorithm for the restoration of a noisy blurred image based on the support vector regression (SVR). Experiments show that the performance of the SVR is very robust in blind image deconvolution where the types of blurs, point spread function (PSF) support, and noise level are all unknown.

To address cost and regulatory concerns, many businesses are converting paper-based elements of their workflows into fully electronic flows that use the content of the documents. Scanning the document contents into workflows, however, is a manual, error-prone, and costly process especially when the data extraction process requires high accuracy. These manual costs are a primary barrier to widespread adoption of distributed capture solutions for business critical workflows such as insurance claims, medical records, or loan applications. Software solutions using artificial intelligence and natural language processing techniques are emerging to address these needs, but each have their individual strengths and weaknesses, and none have demonstrated a high level of accuracy across the many unstructured document types included in these business critical workflows. This paper describes how to overcome many of these limitations by intelligently combining multiple approaches for document classification using meta-algorithmic design patterns. These patterns explore the error space in multiple engines, and provide improved and emergent results in comparison to voting schemes and to the output of any of the individual engines. This paper considers the results of the individual engines along with traditional combinatorial techniques such as voting, before describing prototype results for a variety of novel metaalgorithmic patterns that reduce individual document error rates by up to 13% and reduce system error rates by up to 38%.

The classification of documents with sparse text, and video analysis, relies on accurate image classification. We herein present a method for binary classification that accommodates any number of individual classifiers. Each individual classifier is defined by the critical point between its two means, and its relative weighting is inversely proportional to its expected error rate. Using 10 simple image analysis metrics, we distinguish a set of "natural" and "city" scenes, providing a "semantically meaningful" classification. The optimal combination of 5 of these 10 classifiers provides 85.8% accuracy on a small (120 image) feasibility corpus. When this feasibility corpus is then split into half training and half testing images, the mean accuracy of the optimum set of classifiers was 81.7%. Accuracy as high as 90% was obtained for the test set when training percentage was increased. These results demonstrate that an accurate classifier can be constructed from a large pool of simple classifiers through the use of the statistical ("Normal") classification method described herein.

This paper describes an algorithm for the restoration of a noisy blurred image based on support vector regression. The blind image deconvolution was formulated as a machine learning problem. From the training set, the mapping between the noisy blurred image and the original image are learned by support vector regression (SVR). With the acquired mapping, the degraded image can be restored. Our approach was experimentally compared with the adaptive Lucy-Richardson maximum likelihood (ML) algorithm. In terms of ISNR (improvement of signal to noise ratio), SVR outperforms ML in blind deblurring tests in which the types of blurs, point spread function (PSF) support, and noise energy are all unknown.

Converting paper books into electronic form provides benefits for archiving, distribution and content reuse. However, directly scanned images are usually undesirable for electronic books, and automated content re-mastering is required. In this paper, we describe a comprehensive image processing suite consisting of three major components: 1) image enhancement with deskew, cropping, color correction, contrast enhancement and text sharpening, 2) compound document image compression, and 3) extraction of TOC (table of content) and linking. We built a processing pipeline that automatically converts a set of scanned page images into a high quality and highly compressed e-book in the popular PDF format.

In this paper, we describe an algorithm for identifying a parametrically described blur based on kurtosis minimization. Using different choices for the parameters of the blur, the noisy blurred image is restored using Wiener filter. We use the kurtosis as a measurement of the quality of the restored image. From the set of the candidate deblurred images, the one with the minimum kurtosis is selected. The proposed technique is tested in a simulated experiment on a variety of blurs including atmospheric turbulence blurs, Gaussian blurs, and out-of-focus blurs. The proposed approach is also tested on real blurred images. Moreover, we test the performance when a wrong blur model is given. Our experiments show that the kurtosis minimization measurements match well with methods that maximize PSNR.

Archiving and re-purposing are automated using zoning analysis that performs segmentation (region boundary definition), classification (region typing) and bit-depth determination. For performance throughput reasons, zoning analysis is often performed on a low-resolution (e.g. 50-100 ppi) representation of the document. At these resolutions, heuristic metrics for classification are required. Reported here are metrics for distinguishing photos and color drawings, and a novel classification technique based solely on the statistics of each heuristic metric. The statistical technique allows ready combination of multiple binary classifiers, and provides a lower classification error than simple voting or metric-confidence techniques. This technique permits new metrics to improve the overall classification. The benefit of this technique on archival optimization is shown.

The use of scanners and other capture devices to incorporate film- and paper-based materials into digital workflows is an important part of "digital convergence", or the bringing of paper-based and electronic documents together into the same electronic workflows. The diversity of captured information-from text and mixed-type documents to photos, negatives, slides and transparencies-requires a combination of document analysis techniques to perform, automatically, the segmentation, classification and workflow assignment of the scanned images. We herein present technologies that provide fast (< 1.0 sec) and reliable (> 95% job accuracy) capture solutions for all of these input content types. These solutions offer near real-time capture that provides automated workflow capabilities to a repertoire of scanning hardware: scanners, all-in-one devices, copiers and multifunctional printers. The techniques used to categorize the documents, perform zoning analysis on the documents, and then perform closed loop quality assurance on the documents are presented.

user interface, segmentation, classification, zoning, bottomup analysis, preview display, scanning, click and select Digital capture (scanning in all its forms, and digital photography/video recording), in providing virtually free temporary memory of captured information, allows users to “over-gather” information during capture, and then to discard unwanted material later. For cameras and video recorders, such editing largely consists of discarding images or frames in their entirety. For scanners (and high-resolution camera/video), such editing benefits from a preview capability that provides quick and reliable user interface tools for selecting, filtering and saving specific portions of the input. Appropriate preview user interface (UI) tools ease the accessing, editing and dispatch to desired destination (archive, application, webpage, etc.) of captured information (text, tables, drawings, photos, etc.). In this paper, we present several different means for the user-directed “rapid capture ” of portions of a scanned image. Specifically, we review past, present and future preview-based UI tools that allow efficient and accurate means of capture to the user. The bases of these tools, as described herein, are user-directed zoning analysis, known as “click and select”, which incorporates a bottom-up zoning analysis engine; and statistics-based region classification, which allows rapid reconfiguration of region identification and clustering. We conclude with our view of the future of UI-directed capture.

We present design strategies, implementation preferences and throughput results obtained in deploying a UI-based ground truthing engine as the last step in the quality assurance (QA) for the conversion of a large out-of-print book collection into digital form. A series of automated QA steps were first performed on the document. Five distinct zoning analysis options were deployed and the PDF output thence generated was used to regenerate TIFF files for comparison to the originals. Regenerated TIFFs failing automated QA or a separate visual QA were tagged for ground truthing. Less than 3% of the pages in a 1.2x106-page corpus required ground truthing, resulting in a throughput rate of "fully-proofed" pages of 2x105 pages/man-week. Among the design advantages crucial for this throughput rate was the use of the identical zoning engine for the original production workflow and for the ground truthing engine.

This paper reports emotion recognition results from speech signals, with particular focus on extracting emotion features from the short utterances typical of Interactive Voice Response (IVR) applications. We focus on distinguishing anger versus neutral speech, which is salient to call center applications. We report on classification of other types of emotions such as sadness, boredom, happy, and cold anger. We compare results from using neural networks, Support Vector Machines (SVM), K-Nearest Neighbors, and decision trees. We use a database from the Linguistic Data Consortium at University of Pennsylvania, which is recorded by 8 actors expressing 15 emotions. Results indicate that hot anger and neutral utterances can be distinguished with over 90% accuracy. We show results from recognizing other emotions. We also illustrate which emotions can be clustered together using the selected prosodic features.

Voting is a well-known technique to combine the decisions of peer experts. It is used in fault tolerant applications to mask errors from one or more experts using n-modular redundancy (NMR) and n-version programming. Voting strategies include: majority, weighted voting, plurality; instance runoff voting, threshold voting, and the more general weighted k-out-of-n systems. Before selecting a voting schema for a particular application, we have to understand the various tradeoffs and parameters and how they impact the correctness, reliability, and confidence in the final decision made by a voting system. In this paper, we propose an enumerated simulation approach to automate the behavior analysis of voting schemas with application to majority and plurality voting. We conduct synthetic studies using a simulator that we develop to analyze results from each expert, apply a voting mechanism, and analyze the voting results. The simulator builds a decision tree and uses a depth-first traversal algorithm to obtain the system reliability among other factors. We define and study the following behaviors: 1) the probability of reaching a consensus, "Pc"; 2) reliability of the voting system, "R"; 3) certainly index, "T"; and 4) the confidence index, "C". The parameters controlling the analysis are the number of participating experts, the number of possible output symbols that can be produced by an expert, the probability distribution of each expert's output, and the voting schema. The paper presents an enumerated simulation approach for analyzing voting systems which can be used when the use of theoretical models are challenged by dependencies between experts or uncommon probability distributions of the expert's output.