Recognition of historical documents is still an active research field due to the relatively low recognition accuracy achieved when processing old fonts or low-quality images. In this work, we investigate the use of Large Language Models (LLMs) for the correction of the OCR for old Greek documents. We examine two different old Greek datasets, one machine printed and one typewritten, using a Deep Network based OCR together with several known and easy-to-use LLMs for the correction of the result. Additionally, we synthetically produce erroneous texts and change the LLM prompts in order to further study the behavior of LLMs for correcting old Greek noisy text. Experimental results highlight the potential of LLMs for OCR correction of old Greek documents especially for the cases that the recognition results are relatively poor.

Baseline detection is a simplified text-line extraction that typically serves as pre-processing for Automated Text Recognition. The cBAD competition benchmarks state-of-the-art baseline detection algorithms. It is the successor of cBAD 2017 with a larger dataset that contains more diverse document pages. The images together with the manually annotated groundtruth are made publicly available which allows other teams to benchmark and compare their methods. We could also evaluate the winning method of cBAD 2017 on the newly introduced dataset which now serves as baseline. This competition shows that the performance of automated baseline detection increased substantially since 2017.

Deep convolutional neural networks are today the new baseline for a wide range of machine vision tasks. The problem of keyword spotting is no exception to this rule. Many successful network architectures and learning strategies have been adapted from other vision tasks to create successful keyword spotting systems. In this paper, we argue that various details concerning this adaptation could be re-examined, to the end of building stronger spotting models. In particular, we examine the usefulness of a pyramidal spatial pooling layer versus a simpler approach, and show that a zoning strategy combined with fixed-size inputs can be just as effective while less computationally expensive. We also examine the usefulness of augmentation, class balancing and ensemble learning strategies and propose an improved network. Our hypotheses are tested with numerical experiments on the IAM document collection, where the proposed network outperforms all other existing models.

This tutorial focuses on recent advances and ongoing developments for historical document processing. It includes the main challenges involved, the different tasks that have to be implemented as well as practices and technologies that currently exist in the literature. The focus is given on the most promising techniques, related projects as well as on existing datasets and competitions that can be proved useful to historical document processing research.

The cBAD competition aims at benchmarking state-of-the-art baseline detection algorithms. It is in line with previous competitions such as the ICDAR 2013 Handwriting Segmentation Contest. A new, challenging, dataset was created to test the behavior of state-of-the-art systems on real world data. Since traditional evaluation schemes are not applicable to the size and modality of this dataset, we present a new one that introduces baselines to measure performance. We received submissions from five different teams for both tracks.

The ICDAR 2017 Competition on Historical Document Writer Identification is dedicated to record the most recent advances made in the field of writer identification. The goal of the writer identification task is the retrieval of pages, which have been written by the same author. The test dataset used in this competition consists of 3600 handwritten pages originating from 13th to 20th century. It contains manuscripts from 720 different writers where each writer contributed five pages. This paper describes the dataset, as well as the details of the competition. Five different institutions submitted six methods which were ranked using identification and retrieval metrics. The paper describes the competition details including the dataset, the evaluation measures used as well as a short description of each submitted method.

DIBCO 2017 is the international Competition on Document Image Binarization organized in conjunction with the ICDAR 2017 conference. The general objective of the contest is to identify current advances in document image binarization of machine-printed and handwritten document images using performance evaluation measures that are motivated by document image analysis and recognition requirements. This paper describes the competition details including the evaluation measures used as well as the performance of the 26 submitted methods along with a brief description of each method.

Nonlinear manifold embedding has attracted considerable attention due to its highly-desired property of efficiently encoding local structure, i.e. intrinsic space properties, into a low-dimensional space. The benefit of such an approach is twofold: it leads to compact representations while addressing the often-encountered curse of dimensionality. The latter plays an important role in retrieval applications, such as keyword spotting, where a sorted list of retrieved objects with respect to a distance metric is required. In this work, we explore the efficiency of the popular manifold embedding method t-distributed Stochastic Neighbor Embedding (t-SNE) on the Query-by-Example keyword spotting task. The main contribution of this work is the extension of t-SNE in order to support out-of-sample (OOS) embedding which is essential for mapping query images to the embedding space. The experimental results demonstrate a significant increase in keyword spotting performance when the word similarity is calculated on the embedding space.

In this paper, we propose a novel probabilistic model for lexicon-free handwriting recognition. Model inputs are word images encoded as Pyramidal Histogram Of Character (PHOC) vectors. PHOC vectors have been used as efficient attribute-based, multi-resolution representations of either text strings or word image contents. The proposed model formulates PHOC decoding as the problem of finding the most probable sequence of characters corresponding to the given PHOC. We model PHOC layers as Beta-distributed observations, linked to hidden states that correspond to character estimates. Characters are in turn linked to one another along a Markov chain, encoding language model information. The sequence of characters is estimated using the max-sum algorithm in a process that is akin to Viterbi decoding. Numerical experiments on the well-known George Washington database show competitive recognition results.

Optical Character Recognition (OCR) of ancient Greek polytonic scripts is a challenging task due to the large number of character classes, resulting from variations of diacritical marks on the vowel letters. Classical OCR systems require a character segmentation phase, which in the case of Greek polytonic scripts is the main source of errors that finally affects the overall OCR performance. This paper suggests a character segmentation free HMM-based recognition system and compares its performance with other commercial, open source, and state-of-the art OCR systems. The evaluation has been carried out on a challenging novel dataset of Greek polytonic degraded texts and has shown that HMM-based OCR yields character and word level error rates of 8.61% and 25.30% respectively, which outperforms most of the available OCR systems and it is comparable with the performance of the state-of-the-art system based on LSTM Networks proposed recently.

Word segmentation refers to the process of defining the word regions of a text line. It is a critical stage towards word and character recognition as well as word spotting and mainly concerns three basic stages, namely preprocessing, distance computation and gap classification. In this paper, we propose a novel word segmentation method which uses the Student's-t distribution for the gap classification stage. The main advantage of the Student's-t distribution concerns its robustness to the existence of outliers. In order to test the efficiency of the proposed method we used the four benchmarking datasets of the ICDAR/ICFHR Handwriting Segmentation Contests as well as a historical typewritten dataset of Greek polytonic text. It is observed that the use of mixtures of Student's-t distributions for word segmentation outperforms other gap classification methods in terms of Recognition Accuracy and F-Measure. Also, in terms of all examined benchmarks, the Student's-t is shown to produce a perfect segmentation result in significantly more cases than the state-of-the-art Gaussian mixture model.

Zoning features have been proved one of the most efficient statistical features which provide high speed and low complexity word matching. They are calculated by the density of pixels or pattern characteristics in several zones that the pattern frame is divided. In this paper, an adaptive zoning technique for efficient word spotting is introduced. The main idea is that the zoning features are extracted after cutting the query word in vertical zones, according to its length and pixel distribution along the horizontal axis, and adjusting these boundaries optimally with the corresponding zones in the candidate match-word using Dynamic Time Warping (DTW). This adjustment is performed by coupling every zone of the query word to the corresponding zone of each candidate match-word with the use of the corresponding warping matrix. This process absorbs the ambiguities between the query and the candidate match words and due to this fact it can be applied to both machine-printed and handwritten document images. The proposed word spotting technique is tested using the pixel density as a characteristic feature in every zone and an improvement is recorded compared to other state-of-the-art methods.

The result of a document image segmentation task, e.g. text line or word segmentation, is usually a labeled image with each label corresponding to a different segmented region. For many applications, the segmented regions need to be stored and represented in an efficient way, using simple geometric shapes. A challenging task is to restrict all pixels corresponding to a specific label inside a polygon with a minimum number of vertices. Such a polygon promotes the description simplicity and the storage efficiency, while providing a much more user-friendly representation that can be edited easily. The proposed method is a cost-effective approximation of the minimum-edges polygon problem, computing a contour enclosing only pixels of a certain label and using a greedy algorithm in order to reduce the contour into a minimum-link polygon that retains the separability property between the labeled set of pixels.

In this paper, we present a novel approach for segmentation-based handwritten keyword spotting. The proposed approach relies upon the extraction of a simple yet efficient descriptor which is based on projections of oriented gradients. To this end, a global and a local word image descriptors, together with their combination, are proposed. Retrieval is performed using to the euclidean distance between the descriptors of a query image and the segmented word images. The proposed methods have been evaluated on the dataset of the ICFHR 2014 Competition on handwritten keyword spotting. Experimental results prove the efficiency of the proposed methods compared to several state-of-the-art techniques.

In this paper we propose a new method for automated segmentation of scanned newspaper pages into articles. Article regions are produced as a result of merging sub-article level content and title regions. We use a Bayesian Gaussian mixture model to model page Connected Component information and cluster input into sub-article components. The Bayesian model is conditioned on a prior distribution over region features, aiding classification into titles and content. Using a Dirichlet prior we are able to automatically estimate correctly the number of title and article regions. The method is tested on a dataset of digitized historical newspapers, where visual experimental results are very promising.

Recognition of old Greek document images containing polytonic (multi accent) characters is a challenging task due to the large number of existing character classes (more than 270) which cannot be handled sufficiently by current OCR technologies. Taking into account that the Greek polytonic system was used from the late antiquity until recently, a large amount of scanned Greek documents still remains without full test search capabilities. In order to assist the progress of relevant research, this paper introduces the first publicly available old Greek polytonic database GRPOLY-DB for the evaluation of several document image processing tasks. It contains both machine-printed and handwritten documents as well as annotation with ground-truth information that can be used for training and evaluation of the most commou document image processing tasks, i.e.. text line and word segmentation, test recognition, isolated character recognition and word spotting. Results using several representative baseline technologies are also presented in order to help researchers evaluate their methods and advance the frontiers of old Greek document image recognition and word spotting.

In this paper, we address the problem of word spotting using a shape-based matching scheme between segmented word images represented by local contour features. As in a typical query-by-example (QBE) paradigm, a user selects an instance of the query word from the collection of interest and a ranked list of images is returned, based on their similarity with the query. This is accomplished in two steps. The query image is firstly aligned with the test image according to a similarity measure defined on their descriptors and then the aligned images are matched through a deformable non-rigid point matching algorithm. Experiments are carried out on historical handwritten text, written in Greek and English, respectively. Moreover, comparisons with other QBE methods show the efficiency of our system as well as its flexibility in adapting to different scripts.

In this paper, we present a new approach for off-line isolated character recognition. The proposed method relies upon the application of a projection-based feature extraction stage, which resembles the Radon transform, on both the original image and a set of generated images corresponding to different gradient orientations of the original image. For the classification stage, Support Vectors Machines (SVM) are used. The proposed method is evaluated using one typewritten (GRPOLY-DB - Historical Greek) and two handwritten (CIL - Greek, CEDAR - English) publicly available databases. Experimental results prove the efficiency of the proposed method compared to several state-of-the-art techniques.

Word spotting and recognition are among the most important applications used today in the field of document processing and text understanding. In word spotting, the goal is to search a scanned document for instances of a specific word. In word recognition, we aim to identify the transcription of the document words. While substantial work in both topics has been published, not all are readily adaptible to scripts other than a specific script and/or language. This is especially true for documents written in the polytonic greek script, a script used to write the greek language during a period that approximately spans two millenia. In this work, we extend the attribute-based model for word spotting and recognition recently presented in [1] for use with polytonic greek documents. To this end, we present three alternative ways to extend the model mechanism to handle the greek alphabet and its various combinations of diacritic marks. We have run numerical experiments over polytonic machine-printed and handwritten documents for word spotting and recognition. The proposed model is shown to outperform other state-of-the-art methods in word spotting trials. Regarding polytonic greek unconstrained handwritten word recognition, to the best of our knowledge, this is the first work to address the problem succesfully.

This paper reports on high-performance Optical Character Recognition (OCR) experiments using Long Short-Term Memory (LSTM) Networks for Greek polytonic script. Even though there are many Greek polytonic manuscripts, the digitization of such documents has not been widely applied, and very limited work has been done on the recognition of such scripts. We have collected a large number of diverse document pages of Greek polytonic scripts in a novel database, called Polyton-DB, containing 15; 689 textlines of synthetic and authentic printed scripts and performed baseline experiments using LSTM Networks. Evaluation results show that the character error rate obtained with LSTM varies from 5.51% to 14.68% (depending on the document) and is better than two well-known OCR engines, namely, Tesseract and ABBYY FineReader.

Document image segmentation is a fundamental step in the document image analysis pipeline as it affects the accuracy of subsequent processing steps. An objective and realistic evaluation of page segmentation techniques is crucial for a quantitative comparison among them. In this paper, a goal-oriented performance evaluation methodology that calculates a comprehensive evaluation measure SR (Success Rate) is presented. SR measure reflects the entire performance of a page segmentation technique in a concise quantitative manner. It is a pixel-based approach which avoids the dependence on a strictly defined ground-truth. The proposed evaluation measure SR deals only with text regions and is correlated with the percentage of the text information in which the subsequent processing (e.g. text line segmentation and recognition) can be applied successfully.

Tran Scriptorium is a 3-years project that aims to develop innovative, cost-effective solutions for the indexing, search and full transcription of historical handwritten document images, using Handwritten Text Recognition (HTR) technology. The production of ground-truth (GT) of a dataset of handwritten document images is among the first tasks. We address novel approaches for the faster production of this GT based on crowd-sourcing and on prior-knowledge methods. We also address here a novel low-cost semi-supervised procedure for obtaining pairs of correct line-level aligned detected/extracted text line images and text line transcripts, specially suitable for training models of the HTR technology employed in Tran Scriptorium.

The tranScriptorium project aims to develop innovative, efficient and cost-effective solutions for annotating handwritten historical documents using modern, holistic Handwritten Text Recognition (HTR) technology. Three actions are planned in tranScriptorium: i) improve basic image preprocessing and holistic HTR techniques; ii) develop novel indexing and keyword searching approaches; and iii) capitalize on new, user-friendly interactive-predictive HTR approaches for computer-assisted operation.

Writer identification is important for forensic analysis, helping experts to deliberate on the authenticity of documents. The ICDAR2013 Competition on Writer Identification is part of a competition series (see also ICDAR2011 and ICFHR2012 Writer Identification Contests) which is dedicated to record recent advances in the field of writer identification for Latin scripts using established evaluation performance measures. The benchmarking dataset was created with the help of 250 writers that were asked to copy four parts of text in two Latin based languages (English and Greek). This paper describes the contest details including the evaluation measures used as well as the performance of the 12 submitted methods by 6 different groups along with a short description of each method.

The estimation and correction of handwritten word skew is a difficult and challenging task since it has to be independent of the variations due to handwriting style and writing conditions. In this paper, a coarse-to-fine technique that integrates core-region information is presented. At first, a rough estimation and correction of the skew is accomplished by cutting vertically the word in two overlapping parts, detecting the center of mass in each part and calculating the inclination of the line that connects the two centers of mass. Afterwards, the core-region of the word is detected, the word is cut again in two overlapping parts and the centers of mass are calculated disregarding all the information outside the core-region (ascenders and descenders). The inclination of the line that connects the updated centers of mass corresponds to a finer estimation of the word skew. After correcting the detected skew the last step of core-region detection and skew correction is repeated iteratively in order to reach a finer word skew estimation that will contribute to a successful handwritten word recognition system. Extensive testing based on various test-sets has demonstrated that the proposed method outperforms the state-of-the-art algorithms concerning word skew estimation while it is more robust in variations of the writing style.

The detection and correction of document skew is one of the most important document image analysis steps. The ICDAR2013 Document Image Skew Estimation Contest (DISEC'13) is the first contest which is dedicated to record recent advances in the field of skew estimation using well established evaluation performance measures on a variety of printed document images. The benchmarking dataset that is used contains 1550 images that were obtained from various sources such as newspapers, scientific books and dictionaries. The document images contain figures, tables, diagrams, architectural plans, electrical circuits and they are written in various languages such as English, Chinese and Greek. This paper describes the details of the contest including the evaluation measures used as well as the performance of the twelve methods submitted by ten different groups along with a short description of each method.

DIBCO 2013 is the international Document Image Binarization Contest organized in the context of ICDAR 2013 conference. The general objective of the contest is to identify current advances in document image binarization for both machine-printed and handwritten document images using evaluation performance measures that conform to document image analysis and recognition. This paper describes the contest details including the evaluation measures used as well as the performance of the 23 submitted methods along with a short description of each method.

This paper presents the results of the Handwriting Segmentation Contest that was organized in the context of the ICDAR2013. The general objective of the contest was to use well established evaluation practices and procedures to record recent advances in off-line handwriting segmentation. Two benchmarking datasets, one for text line and one for word segmentation, were created in order to test and compare all submitted algorithms as well as some state-of-the-art methods for handwritten document image segmentation in realistic circumstances. Handwritten document images were produced by many writers in two Latin based languages (English and Greek) and in one Indian language (Bangla, the second most popular language in India). These images were manually annotated in order to produce the ground truth which corresponds to the correct text line and word segmentation results. The datasets of previously organized contests (ICDAR2007, ICDAR2009 and ICFHR2010 Handwriting Segmentation Contests) along with a dataset of Bangla document images were used as training dataset. Eleven methods are submitted in this competition. A brief description of the submitted algorithms, the evaluation criteria and the segmentation results obtained from the submitted methods are also provided in this manuscript.

Word retrieval is an important task in the area of document analysis and recognition. The selection of appropriate features is a crucial step in the word matching and retrieval process. Several efficient techniques have been proposed which use a wide range of features. This paper proposes a methodology for the efficient fusion of multiple ranking results produced by different word matching techniques. Specifically, a Minimum Ranking method is proposed for the combination of two or more ranking results. The method is compared with two state-of-the-art ranking fusion methods. The experimental results show that the fusion of the ranked results outperforms the ranking efficiency of the individual systems. Moreover, the proposed Minimum Ranking method outperforms the other two state-of-the-art fusion methods.

In this paper, we propose a new technique for estimating the word slant that is based on analyzing the non-horizontal parts of the characters. We calculate the orientation, the height of the bounding box that includes all non-horizontal parts as well as their location related to the word. We estimate the non-horizontal parts orientation weighted according to the height of the corresponding bounding box. An additional weight is applied if the fragment is outside of the core-region of the word, which indicates that this fragment is probably one of the strokes that should be, by definition, vertical to text orientation. Extensive experimental results prove the efficiency of the proposed method.

In this paper, we propose a word recognition methodology based on a novel size-normalization and a pixel-based image dissimilarity measure. As a first step, we apply a new size-normalization technique using baseline estimation. Starting from those size-normalized images, the difference between two word images is calculated using an image dissimilarity measure based on curvature estimation using integral invariants and a windowed Hausorff distance. We conducted several experiments comparing the new methodology with state-of-the-art techniques using ground truth data from a historical book. The experiments prove the efficiency of the proposed size normalization as well as of the overall proposed sytem.

Zoning features are of the most popular and efficient statistical features that provide high speed and low complexity for character and word recognition. They are calculated by the density of pixels or pattern characteristics in several zones we divide the pattern frame. In this paper, we introduce the idea of adaptive zoning features that are extracted after adjusting the position of every zone based on local pattern information. This adjustment is performed by moving every zone towards the pattern body. This process is based on the maximization of the local pixel density around each zone. We have extensively tested our approach for character and word recognition as well as for using the pixel density or pattern characteristics in every zone. For all cases, we have recorded a significant improvement when the zoning features are used in the proposed adaptive way.

Recognition of document images having Greek polytonic (multi accent) characters is a challenging task due the large number of existing character classes (more than 270). In this paper, we propose a novel OCR framework for the recognition of machine-printed Greek polytonic documents that is based on combining five different recognition modules in order to have a small number of classes (around 30) in each module. One recognition module is used for accent recognition while four recognition modules are used for the recognition of characters belonging to different horizontal text zones. The proposed system also includes the following stages: (a) pre-processing, (b) text dewarping, text line and text baseline detection, (c) accent and character detection and (d) combination of accent and character recognition results. Extended experiments have been conducted in order to record the performance of the proposed OCR system, of all involved recognition modules as well as of the accent detection stage.

Word spotting is an alternative methodology for document indexing based on spotting words directly on document images with the help of efficient word matching while avoiding conventional OCR procedure. The result of the word spotting procedure is a list of word images ranked according to a certain similarity criterion. In this paper, we propose an efficient method to cut-off the ranked list in order to provide the best tradeoff between recall and precision rates. Our aim is to filter the most relevant results based on a threshold which corresponds to an approximate maximization of the expected F-Measure. This is achieved by introducing an estimator that combines the distance of each ranked word with its cumulative moving average. Experimental results on a database with representative historical printed documents prove the efficiency of the proposed approach.

ICDAR 2011 Writer Identification Contest is the first contest which is dedicated to record recent advances in the field of writer identification using established evaluation performance measures. The benchmarking dataset of the contest was created with the help of 26 writers that were asked to copy eight pages that contain text in several languages (English, French, German and Greek). This paper describes the contest details including the evaluation measures used as well as the performance of the 8 submitted methods along with a short description of each method.

In this paper we present a binarization technique for textual content in video frames which can be applied in the resulting image of the text detection step aiming in an improved OCR performance. The proposed technique is based on the detection of the text baselines in order to define the main body of the text. The main body of the text is used to detect the stroke width of the characters which will address the two consecutive locally adaptive binarization steps that follow. At the first step, we use different valuation in parameters for the inside and outside area of the main body of the text. To include the thinned or broken binarized parts that may exist outside the main text body, convex hull analysis is performed so that the entire text body is considered. At the second step, binarization is performed with different valuation in parameters for the inside and outside area of the entire text body. The effectiveness of the proposed technique is demonstrated by both qualitative and OCR-based evaluation.

Document skew detection is often done by the use of horizontal projections. In this paper, we introduce a new document skew detection approach that is based on vertical projections as well as bounding box minimization criterion. We motivated by the fact that the majority of the Latin characters have vertical strokes. We claim that the proposed approach is more efficient and gives more accurate results compared with the state-of-the-art skew detection algorithm based on horizontal projections since it is noise and warp resistant. For these reasons, it can be efficiently applied to historical machine printed documents. Experimental results on a database with representative historical printed documents prove the efficiency of the proposed approach. Moreover, an analysis of the performance of the main elements of the proposed technique is also done.

DIBCO 2011 is the International Document Image Binarization Contest organized in the context of ICDAR 2011 conference. The general objective of the contest is to identify current advances in document image binarization for both machine-printed and handwritten document images using evaluation performance measures that conform to document image analysis and recognition. This paper describes the contest details including the evaluation measures used as well as the performance of the 18 submitted methods along with a short description of each method.

Scanning two book pages at the same time helps to accelerate the scanning process but on the other hand introduces several difficulties if the user needs to have one page per image. A major difficulty is the appearance of noisy black borders around text areas as well as of noisy black stripes between the two pages. In this paper, we propose a novel algorithm for detecting the page frames on double page document images. Our aim is to split the image into the two pages as well as to remove noisy borders. First we apply a pre-processing which includes binarization, noise removal and image smoothing. Then, we detect the vertical zones of the two pages. In this stage, we introduce the vertical white run projections which have been proved efficient for detecting vertical zones of text areas. Finally, the horizontal zones of the two pages are detected based on horizontal white run projections. The experimental results on several double page document images from fifteen different books demonstrate the effectiveness of the proposed technique.

A major difficulty for designing a document image segmentation methodology is the proper value selection for all involved parameters. This is usually done after experimentations or after involving a training supervised phase which is a tedious process since the corresponding segmentation ground truth has to be created. In this paper, we propose a novel automatic unsupervised parameter selection methodology that can be applied to the character segmentation problem. It is based on clustering of the entities obtained as a result of the segmentation for different values of the parameters involved in the segmentation method. The clustering is performed using features extracted from the segmented entities based on zones and from the area that is formed from the projections of the upper/lower and left/right profiles. Optimization of an appropriate intra-class distance measure yields the optimal parameter vector. The method is evaluated on two segmentation algorithms, namely a recently proposed character segmentation technique based on skeleton segmentation paths, as well as the well known RLSA technique. The proposed parameter selection method is capable of finding the segmentation parameters that correspond to the optimal or near optimal segmentation result, as this is determined by counting the number of matches between the entities detected by the segmentation algorithm and the entities in the ground truth.

DIBCO 2009 is the first International Document Image Binarization Contest organized in the context of ICDAR 2009 conference. The general objective of the contest is to identify current advances in document image binarization using established evaluation performance measures. This paper describes the contest details including the evaluation measures used as well as the performance of the 43 submitted methods along with a short description of each method.

In this paper, a new efficient word spotting methodology is presented that can be applied to historical printed documents without requiring any previous block or word segmentation step. Our aim is to address a methodology which is segmentation-free since in many cases of historical documents, the segmentation process does not produce meaningful results due to unconstraint layout, several degradations or typesetting imperfections. The proposed method is based on block-based document image descriptors that are used at a template matching process satisfying invariance in terms of translation, rotation and scaling. Improvement in terms of time expense is obtained by applying the matching process only on salient regions of the image. Experimental results on a database with representative historical printed documents prove the efficiency of the proposed approach.

The Handwriting Segmentation Contest was organized in the context of ICDAR2009 conference in order to record recent advances in off-line handwriting segmentation. This paper describes the contest details including the dataset, the ground truth and the evaluation criteria and presents the results of the 12 participating methods. The contest includes handwritten document images produced by many writers in several languages (English, French, German and Greek). These images are manually annotated in order to produce the ground truth which corresponds to the correct text line and word segmentation result. For the evaluation, a well established approach is used based on counting the number of matches between the entities detected by the segmentation algorithm and the entities in the ground truth.

Word segmentation is a critical stage towards word and character recognition as well as word spotting and mainly concerns two basic aspects, distance computation and gap classification. In this paper, we propose a robust evaluation methodology that treats the distance computation and the gap classification stages independently. The detection rate calculated for every distance metric corresponds to the maximum detection rate that we could have achieved if we had a perfect classifier for the gap classification stage. The proposed evaluation framework has been applied to several state-of-the-art techniques using a handwritten as well as a historical typewritten document set. The best combination of distance metric computation and gap classification state-of-the-art techniques is proposed.

In this paper, we propose a novel technique to segment handwritten document images into text lines by shredding their surface with local minima tracers. Our approach is based on the topological assumption that for each text line, there exists a path from one side of the image to the other that traverses only one text line. We first blur the image and then use tracers to follow the white-most and black-most paths from left to right as well as from right to left in order to shred the image into text line areas. We experimentally tested the proposed methodology and got promising results comparable to state of the art text line segmentation techniques.

In this paper, a new document image binarization technique is presented, as an improved version of the state-of-the-art adaptive logical level technique (ALLT). The original ALLT depends on fixed windows to extract essential features such as the character stroke width. Since characters with several different stroke widths may exist within a region, this can lead to erroneous results. In our approach, we use local adaptive binarization as a guide to our adaptive stroke width detection. The skeleton and the contour points of the binarization output are combined to identify locally the stroke width. Additionally, we introduce an adaptive local parameter ldquobetardquo that enhances the characters and improves the overall performance. In this way, we achieve more accurate binarization results in both handwritten and printed documents with a particular focus on degraded historical documents. Experimental results prove the effectiveness of the proposed technique compared to other state-of-the-art methodologies.

One of the major challenges in camera document analysis is to deal with the page curl and perspective distortions. In spite of the prevalence of dewarping techniques, no standard for their performance evaluation method exists with most of the evaluation done to concentrate in visual pleasing impressions. This paper presents an objective evaluation methodology for document image dewarping techniques. First, manually selected sets of points of the initial warped image are matched with the corresponding points of the dewarping result using the scale invariant feature transform (SIFT). Each set corresponds to a representative text line of the image. Then, based on cubic polynomial curves that fit to the selected text lines, a comprehensive measure which reflects the entire performance of a dewarping technique in a concise quantitative manner is calculated. Experiments applying the proposed performance evaluation methodology on two state of the art dewarping techniques as well as a commercial package are presented.

In this paper, we present a methodology for off-line character recognition that mainly focuses on handling the difficult cases of historical fonts and styles. The proposed methodology relies on a new feature extraction technique based on recursive subdivisions of the image as well as on calculation of the centre of masses of each sub-image with sub-pixel accuracy. Feature extraction is followed by a hierarchical classification scheme based on the level of granularity of the feature extraction method. Pairs of classes with high values in the confusion matrix are merged at a certain level and higher level granularity features are employed for distinguishing them. Several historical documents were used in order to demonstrate the efficiency of the proposed technique.

This paper proposes a hybrid system for text detection in video frames. The system consists of two main stages. In the first stage text regions are detected based on the edge map of the image leading in a high recall rate with minimum computation requirements. In the sequel, a refinement stage uses an SVM classifier trained on features obtained by a new local binary pattern based operator which results in diminishing false alarms. Experimental results show the overall performance of the system that proves the discriminating ability of the proposed feature set.

This paper presents a new adaptive approach for the binarization and enhancement of historical and degraded documents. The proposed method is based on (i) efficient pre-processing; (ii) the combination of the results of several state-of-the-art binarization methodologies; (iii) the incorporation of edge information and (iv) the application of efficient image post-processing based on mathematical morphology for the enhancement of the final result. The proposed method demonstrated superior performance against six well-known techniques on numerous historical handwritten and machine-printed documents mainly from the Library of Congress of the United States archive. The performance evaluation was based on a consistent and concrete methodology.

In this paper we propose a novel and efficient technique for finding keywords typed by the user in digitised machine-printed historical documents using the dynamic time warping (DTW) algorithm. The method uses word portions located at the beginning and end of each segmented word of the processed documents and try to estimate the position of the first and last characters in order to reduce the list of candidate words. Since DTW can become computational intensive in large datasets the proposed method manages to significantly prune the list of candidate words thus, speeding up the entire process. Word length is also used as a means of further reducing the data to be processed. Results are improved in terms of time and efficiency compared to those produced if no pruning is done to the list of candidate words.

Evaluation of document image binarization techniques is a tedious task that is mainly performedby a human expert or by involving an OCR engine. This paper presents an objective evaluation methodology for document image binarization techniques that aims to reduce the human involvement in the ground truth construction and consecutive testing. A skeletonized ground truth image is produced by the user following a semi-automatic procedure. The estimated ground truth image can aid in evaluating the binarization result in terms of recall and precision as well as to further analyze the result by calculating broken and missing text, deformations and false alarms. A detailed description of the methodology along with a benchmarking of the six (6) most promising state-of-the-art binarization algorithms based on the proposed methodology is presented.

Dewarping of camera document images has attracted a lot of interest over the last few years since warping not only reduces the document readability but also affects the accuracy of an OCR application. In this paper, a two-step approach for efficient dewarping of camera document images is presented. At a first step, a coarse dewarping is accomplished with the help of a transformation model which maps the projection of a curved surface to a 2D rectangular area. The projection of the curved surface is delimited by the two curved lines which fit the top and bottom text lines along with the two straight lines which fit to the left and right text boundaries. At a second step, fine dewarping is achieved based on words detection. All words are pose normalized guided by the lower and upper word baselines. Experimental results on several camera document images demonstrate the robustness and effectiveness of the proposed technique.

In this paper a complete OCR methodology for recognizing historical documents, either printed or handwritten without any knowledge of the font, is presented. This methodology consists of three steps: The first two steps refer to creating a database for training using a set of documents, while the third one refers to recognition of new document images. First, a pre-processing step that includes image binarization and enhancement takes place. At a second step a top-down segmentation approach is used in order to detect text lines, words and characters. A clustering scheme is then adopted in order to group characters of similar shape. This is a semi-automatic procedure since the user is able to interact at any time in order to correct possible errors of clustering and assign an ASCII label. After this step, a database is created in order to be used for recognition. Finally, in the third step, for every new document image the above segmentation approach takes place while the recognition is based on the character database that has been produced at the previous step.

This paper continues the authors' attempt to address the need for objective comparative evaluation of layout analysis methods in realistic circumstances. It describes the Page Segmentation Competition (modus operandi, dataset and evaluation criteria) held in the context of ICDAR2007 and presents the results of the evaluation of three candidate methods. The main objective of the competition was to compare the performance of such methods using scanned documents from commonly-occurring publications. The results indicate that although methods continue to mature, there is still a considerable need to develop robust methods that deal with everyday documents.

This paper presents the results of the handwriting segmentation contest that was organized in the context of ICDAR2007. The aim of this contest was to use well established evaluation practices and procedures in order to record recent advances in off-line handwriting segmentation. Two benchmarking datasets (one for text line and one for word segmentation) were used in a common evaluation platform in order to test and compare all submitted algorithms for handwritten document segmentation in realistic circumstances. The results of the evaluation of five algorithms submitted by participants as well as of two state-of-the-art algorithms are presented. The performance evaluation method is based on counting the number of matches between the text lines or words detected by the algorithms and the text line or words of the ground truth.

Non-linear warping appears in document images when captured by a digital camera or a scanner, especially in the case that these documents are digitized bounded volumes. Arbitrarily warped documents may have several slope changes along the text lines as well as along the words of the same text line. In this paper, a novel segmentation based technique for efficient restoration of arbitrarily warped document images is presented. The proposed technique recovers the documents relying upon (i) text lines and words detection using a novel segmentation technique appropriate for warped documents, (ii) a first draft binary image de-warping based on word rotation and translation according to upper and lower word baselines, and (Hi) a recovery of the original warped image guided by the draft binary image de-warping result. Experimental results on several arbitrarily warped documents prove the effectiveness of the proposed technique.

In this paper we present a new text line detection method for unconstrained handwritten documents. The proposed technique is based on a strategy that consists of three distinct steps. The first step includes image binarization and enhancement, connected component extraction and average character height estimation. In the second step, a block-based Hough transform is used for the detection of potential text lines while a third step is used to correct possible splitting, to detect text lines that the previous step did not reveal and, finally, to separate vertically connected characters and assign them to text lines. The performance evaluation of the proposed approach is based on a consistent and concrete evaluation methodology.

Word segmentation is a crucial step for segmentation-free document analysis systems and is used for creating an index based on word matching. In this paper, we propose a novel methodology for word segmentation in historical and degraded machine-printed documents. The proposed technique faces problems such as having text of different size, having text and non-text areas lying very near and having non-straight and warped text lines. It is based on: (i) a dynamic run length smoothing algorithm that helps grouping together homogeneous text regions, (ii) noise and punctuation marks removal as well as on obstacle detection in order to facilitate the segmentation process and (iv) a draft text line estimation procedure that guides the final word segmentation result. After testing on numerous historical and degraded machine-printed documents, it has turned out that our methodology performs better compared to current state-of-the-art word segmentation techniques for historical and degraded machine-printed documents.

In this paper, we present an off-line methodology for isolated Greek handwritten character recognition based on efficient feature extraction followed by a suitable feature vector dimensionality reduction scheme. Extracted features are based on (i) horizontal and vertical zones, (ii) the projections of the character profiles, (Hi) distances from the character boundaries and (iv) profiles from the character edges. The combination of these types of features leads to a 325- dimensional feature vector. At a next step, a dimensionality reduction technique is applied, according to which the dimension of the feature space is lowered down to comprise only the features pertinent to the discrimination of characters into the given set of letters. In this paper, we also present a new Greek handwritten database of 36,960 characters that we created in order to measure the performance of the proposed methodology.

There is an established need for objective evaluation of layout analysis methods, in realistic circumstances. This paper describes the page segmentation competition (modus operandi, dataset and evaluation criteria) held in the context of ICDAR2005 and presents the results of the evaluation of four candidate methods. The main objective of the competition was to compare the performance of such methods using scanned documents from commonly-occurring publications. The results indicate that although methods seem to be maturing, there is still a considerable need to develop robust methods that deal with everyday documents.

In this paper, we propose a novel segmentation-free approach for keyword search in historical typewritten documents combining image preprocessing, synthetic data creation, word spotting and user feedback technologies. Our aim is to search for keywords typed by the user in a large collection of digitized typewritten historical documents. The proposed method is based on: (i) image preprocessing for image binarization and enhancement, noisy border and frame removal, orientation and skew correction; (ii) creation of synthetic image words from keywords typed by the user; (Hi) word segmentation using dynamic parameters; (iv) efficient feature extraction for each image word and (v) a retrieval procedure that is optimized by user's feedback. Experimental results prove the efficiency of the proposed approach.

Recognition of handwritten manuscripts is essential for efficient content exploitation of the valuable old Greek historical collections. In this paper, we focus on the problem of recognizing old Greek handwritten manuscripts and propose a novel recognition technique that can be applied to a large number of important historical manuscript collections which are written in lower case letters and originate from St. Catherine's Mount Sinai Monastery. Based on an open and closed cavity character representation, we propose a novel, segmentation-free, fast and efficient technique for the detection and recognition of characters and character ligatures. First, we detect open and closed cavities that exist in the skeletonized character body. Then, the recognition of a specific character or character ligature is based on the protrusible segments that appear in the topological description of the character skeletons. Experimental results prove the efficiency of the proposed approach.

This paper presents the results of the First International Newspaper Segmentation contest that was organized on the frame of ICDAR 2001 conference. The aim of this contest was to evaluate all existing algorithms for document image segmentation that can be applied to Newspaper page segmentation. We evaluated the performance of three different newspaper segmentation algorithms on tracing all basic entities that appear in newspaper pages from the beginning of the previous century up to the present. The selected entities are text regions, lines and images/drawings. Both training and test sets come from Greek and English newspapers. The performance evaluation method is based on counting the number of matches between the entities detected by the algorithms and the entities of the ground truth. In order to rank the global performance of each participant, we employed a metric that combines the average values of detection rate and recognition accuracy.

Run length smoothing algorithm (RLSA) and projection profiles are among the fundamental algorithms in binary image processing, mainly used for segmentation of monochrome images. In this paper, fast RLSA and projection profiles are applied to binary images represented by a set of nonoverlapping rectangular blocks. The representation of binary images using rectangular blocks as primitives has been used with great success for several image processing tasks, such as image compression, Hough transform fast implementation and skeletonization. We show that this representation can be applied with great success for fast RLSA application and fast projection profiles evaluation. The experimental results demonstrate that starting from a block represented binary image we can apply RLSA and evaluate projection profiles in significant less CPU time. The average time gain is recorded at 60% and 88%, respectively.

No abstract available.

The conversion of newspaper pages into digital resources is an important task that greatly contributes to the preservation of and access to newspaper archives. In this paper, an integrated methodology is presented for segmenting newspaper pages and identifying newspaper articles. In the first stage, a succession of image processing and document analysis algorithms is employed for segmenting newspaper page images into various objects (text, images and drawings, titles). A rule based approach is subsequently applied to the objects identified during the page segmentation phase for reconstructing individual articles. Experimental results, obtained from a large testbed of old newspaper issues, are presented which clearly demonstrate the applicability of our integrated approach to successful newspaper page segmentation and identification of newspaper articles.