Search

Parkinson’s disease (PD) is a complex disorder characterized by several motor and non-motor symptoms that worsen over time, and that differ from person to another. In the early stages, when the symptoms are often incomplete, the diagnosis becomes difficult and at times, the subject may remain undiagnosed. This difficulty is a strong motivation for computer-based assessment tools that can aid in the early diagnosing and predicting the progression of PD. Handwriting’s deterioration, vocal and eye movement impairments may be ones of the earliest indicators for the onset of the illness. A language independent model to detect PD at early stages by using multimodal signals has not been enough addressed. Due to the lack of multimodal and multilingual databases, database which includes online handwriting, speech signals, and eye movement’s recordings have been recently collected. After succeeding in building language independent models for PD early diagnosis using pure handwriting or speech, we propose in this work language independent models based on bimodal analyses (handwriting and speech), where both SVM and deep learning models are studied. Our experiments show that classification accuracy up to 100% can be obtained by our SVM model through handwriting/speech bimodal analysis.

Degenerative disorders such as Parkinson’s disease (PD) have an influence on daily activities due to rigidity of muscles, tremor or cognitive impairment. Micrographia, speech intensity, and deficient generation of voluntary saccadic eye movements (Pretegiani and Optican in Front Neurol 8:592, 2017) are manifestations of PD that can be used to devise noninvasive and low cost clinical tests. In this context, we have collected a multimodal dataset that we call Parkinson’s disease Multi-Modal Collection (PDMultiMC), which includes online handwriting, speech signals, and eye movements recordings. We present here the handwriting dataset that we call HandPDMultiMC that will be made publicly available. The HandPDMultiMC dataset includes handwriting samples from 42 subjects (21 PD and 21 controls). In this work we investigate the application of various Deep learning architectures, namely the CNN and the CNN-BLSTM, to PD detection through time series classification. Various approaches such as Spectrograms have been applied to encode pen-based signals into images for the CNN model, while the raw time series are directly used in the CNN-BLSTM. In order to train these models for PD detection on large scale data, various data augmentation approaches for pen-based signals are proposed. Experimental results on our dataset show that the best performance for early PD detection (97.62% accuracy) is reached by a combination of CNN-BLSTM models trained with Jittering and Synthetic data augmentation approaches. We also illustrate that deep architectures can surpass the models trained on pre-engineered features even though the available data is small.

Deep learning has been successfully applied to different classification applications where large data are available. However, the lack of data makes it more difficult to predict Parkinson’s disease (PD) with the deep models, which requires enough number of training data. Online handwriting dynamic signals can provide more detailed and complex information for PD detection task. In our previous work [1], two different deep models were studied for time series classification; the convolutional neural network (CNN) and the convolutional neural network- bidirectional long short term memory network (CNN-BLSTM). Different approaches were applied to encode pen-based signals into images for the CNN model while the raw time series are used directly with the CNN-BLSTM model. We have showed that both CNN model with spectrogram images as input and CNN-BLSTM model, improve the performance of time series classification applied for early PD stage detection. However, these approaches did not outperform classical support vector machine (SVM) classification applied on pre-engineered features. In this paper we investigate transfer learning and data augmentation approaches in order to train these models for PD detection on large-scale data. Various data augmentation methods for pen-based signals are proposed. Our experimental results show that the CNN-BLSTM model used with the combination of Jittering and Synthetic data augmentation methods provides promising results in the context of early PD detection, with accuracy reaching 97.62%. We have illustrated that deep architecture can surpass the models trained on pre-engineered features even though the available data is small.

Parkinson's disease (PD) is a neurological disorder associated with a progressive decline in motor skills, speech, and cognitive processes. Since the diagnosis of Parkinson's disease is difficult, researchers have worked to develop a support tool based on algorithms to separate healthy controls from PD patients. Online handwriting dynamic signals can provide more detailed and complex information for PD detection task. Existing techniques often depended on hand-crafted features that required expert knowledge of the field. In this paper, it is suggested to learn pen-based features by means of deep learning for automatic classification of PD. For this purpose, a visual representation of the time series can be computed and used at the input of a convolutional neural network (CNN) as in [4]. Classically, the time series is transformed into a fixed dimension image applying normalization on the time dimension. In this work we have experimented several visual representations, including the spectrogram where normalization of the time scale is applied after short term information has been extracted locally. We have been able to show that considering the local short term information allows the deep learning models to provide better classification results compared to a globally normalized fixed dimension visual representation. For validation purpose, a CNN-BLSTM was directly applied on the time series, without any normalization of the time scale which led to best performance equivalent to the one obtained on spectrogram representation.

Nowadays, governmental programs like ESA’s Copernicus provide freely available data that can be easily utilized for earth observation. In the present work, the problem of detecting agricultural and non-agricultural land cover is addressed. The methodology is based on classification with convolutional neural networks (CNNs) and transfer learning using AlexNet. The study area is located at the Ionian Islands, which include several land cover classes according to Copernicus CORINE Land Cover 2018 (CLC 2018). Furthermore, the dataset consists of natural color images acquired by Sentinel-2A multi-spectral instrument. Experimentation proves that extra addition of training data from foreign grounds, unfamiliar to the Greek data, serves much as a confusing agent regarding network performance.

Slanted text has been demonstrated to be a salient feature of handwriting. Its estimation is a necessary preprocessing task in many document image processing systems in order to improve the required training. This paper describes and evaluates a new technique for removing the slant from historical document pages that avoids the segmentation procedure into text lines and words. The proposed technique first relies on slant angle detection from an accurate selection of fragments. Then, a slant removal technique is applied. However, the presented slant removal technique may be combined with any other slant detection algorithm. Experimental results are provided for four document image databases: two historical document databases, the TrigraphSlant database (the only database dedicated to slant removal), and a printed database in order to check the precision of the proposed technique.

A reliable system depending on algorithms that assist in the decision-making process to diagnose Parkinson’s disease (PD) at an early stage and to predict the Hoehn & Yahr (H&Y) stage and the unified Parkinson’s disease rating scale (UPDRS) score is developed. In a previous work [3], we used features extracted from Arabic handwriting for diagnosing PD as binary decision. In this work, we use these features for constructing a prediction model that evaluates the H&Y stage and the UPDRS scores. A multi-class support vector machine (SVM) classifier is trained using re-sampling approaches such as adaptive synthetic sampling approach (ADASYN). The classifier is evaluated with 4-fold cross validation. The experiments show that H&Y stage, UPDRS scores, and total UPDRS can be predicted with accuracies of 94%, 92%, and 88% respectively. The proposed method can be implemented as an efficient clinical decision support system for early detection and monitoring the progression of PD.

Historical documents present many challenges for offline handwriting recognition systems, among them, the segmentation and labeling steps. Carefully annotated text lines are needed to train an HTR system. In some scenarios, transcripts are only available at the paragraph level with no text-line information. In this work, we demonstrate how to train an HTR system with few labeled data. Specifically, we train a deep convolutional recurrent neural network (CRNN) system on only 10% of manually labeled text-line data from a dataset and propose an incremental training procedure that covers the rest of the data. Performance is further increased by augmenting the training set with specially crafted multi scale data. We also propose a model-based normalization scheme which considers the variability in the writing scale at the recognition phase. We apply this approach to the publicly available READ dataset. Our system achieved the second best result during the ICDAR2017 competition [1].

[EN] The digitization of historical handwritten document images is important for the preservation of cultural heritage. Moreover, the transcription of text images obtained from digitization is necessary to provide efficient information access to the content of these documents. Handwritten Text Recognition (HTR) has become an important research topic in the areas of image and computational language processing that allows us to obtain transcriptions from text images. State-of-the-art HTR systems are, however, far from perfect. One difficulty is that they have to cope with image noise and handwriting variability. Another difficulty is the presence of a large amount of Out-Of-Vocabulary (OOV) words in ancient historical texts. A solution to this problem is to use external lexical resources, but such resources might be scarce or unavailable given the nature and the age of such documents. This work proposes a solution to avoid this limitation. It consists of associating a powerful optical recognition system that will cope with image noise and variability, with a language model based on sub-lexical units that will model OOV words. Such a language modeling approach reduces the size of the lexicon while increasing the lexicon coverage. Experiments are first conducted on the publicly available Rodrigo dataset, which contains the digitization of an ancient Spanish manuscript, with a recognizer based on Hidden Markov Models (HMMs). They show that sub-lexical units outperform word units in terms of Word Error Rate (WER), Character Error Rate (CER) and OOV word accuracy rate. This approach is then applied to deep net classifiers, namely Bi-directional Long-Short Term Memory (BLSTMs) and Convolutional Recurrent Neural Nets (CRNNs). Results show that CRNNs outperform HMMs and BLSTMs, reaching the lowest WER and CER for this image dataset and significantly improving OOV recognition., Work partially supported by projects READ: Recognition and Enrichment of Archival Documents - 674943 (European Union's H2020) and CoMUN-HaT: Context, Multimodality and User Collaboration in Handwritten Text Processing - TIN2015-70924-C2-1-R (MINECO/FEDER), and a DGA-MRIS (Direction Generale de l'Armement - Mission pour la Recherche et l'Innovation Scientifique) scholarship.

International audience; Parkinson's disease (PD) is a neurological disorder associated with a progressive decline in motor skills, speech, and cognitive processes. Since the diagnosis of Parkinson's disease is difficult, researchers have worked to develop a support tool based on algorithms to differentiate healthy controls from PD patients. Online handwriting analysis is one of the methods that can be used to diagnose PD. The aim of this study is to find a subset of handwriting features suitable for efficiently identifying subjects with PD. Data was taken from PDMultiMC database collected in Lebanon, and consisting of 16 medicated PD patients and 16 age matched controls. Seven handwriting tasks were collected such as copying patterns, copying words in Arabic, and writing full names. For each task kinematic and spatio-temporal, pressure, energy, entropy, and intrinsic features were extracted. Feature selection was done in two stages, the first stage selected a subset using statistical analysis, and the second step select the most relevant features of this subset, by a suboptimal approach. The selected features were fed to a support vector machine classifier with RBF kernel, whose aim is to identify the subjects suffering from PD. The accuracy of the classification of PD was as high as 96.875%, with sensitivity and specificity equal to 93.75 % and 100%. The results as well as the selected features suggest that handwriting can be a valuable marker as a diagnosis tool.

This paper reports a new approach based on convolutional neural networks (CNNs), which uses spatial transformer networks (STNs). The approach, referred to as Tied Spatial Transformer Networks (TSTNs), consists of training a system which combines a localization CNN and a classification CNN whose weights are shared. The localization CNN is used for predicting an affine transform for the input image, which is then processed according to the predicted parameters and passed through the classification CNN. We have conducted initial experiments on the cluttered MNIST dataset of noisy digits, comparing the TSTN and STN with identical configurations of trainable parameters, but untied, as well as the classification CNN only, applied to the unprocessed images. In all these cases, we obtain better results using the TSTN. We conjecture that the TSTN provides a regularization effect, as compared to untied STNs. Further experiments seem to support this hypothesis.

International audience

This paper proposes a novel method for document enhancement which combines two recent powerful noise-reduction steps. The first step is based on the Total Variation framework. It flattens background grey-levels and produces an intermediate image where background noise is considerably reduced. This image is used as a mask to produce an image with a cleaner background while keeping character details. The second step is applied to the cleaner image and consists of a filter based on Non-local Means: character edges are smoothed by searching for similar patch images in pixel neighborhoods. The document images to be enhanced are real historical printed documents from several periods which include several defects in their background and on character edges. These defects result from scanning, paper aging and bleed-through. The proposed method enhances document images by combining the Total Variation and the Non-local Means techniques in order to improve OCR recognition. The method is shown to be more powerful than when these techniques are used alone and than other enhancement methods.

The problem addressed in this study is the offline recognition of handwritten Arabic city names. The names are assumed to belong to a fixed lexicon of about 1,000 entries. A state-of-the-art classical right-left hidden Markov model (HMM)-based recognizer (reference system) using the sliding window approach is developed. The feature set includes both baseline-independent and baseline-dependent features. The analysis of the errors made by the recognizer shows that the inclination, overlap, and shifted positions of diacritical marks are major sources of errors. In this paper, we propose coping with these problems. Our approach relies on the combination of three homogeneous HMM-based classifiers. All classifiers have the same topology as the reference system and differ only in the orientation of the sliding window. We compare three combination schemes of these classifiers at the decision level. Our reported results on the benchmark IFN/ENIT database of Arabic Tunisian city names give a recognition rate higher than 90 percent accuracy and demonstrate the superiority of the neural network-based combination. Our results also show that the combination of classifiers performs better than a single classifier dealing with slant-corrected images and that the approach is robust for a wide range of orientation angles.

In this paper, we investigate the application of dynamic Bayesian networks (DBNs) to the recognition of degraded characters. DBNs are an extension of one-dimensional hidden Markov models (HMMs) which can handle several observation and state sequences. In our study, characters are represented by the coupling of two HMM architectures into a single DBN model. The interacting HMMs are a vertical HMM and a horizontal HMM whose observable outputs are the image columns and image rows, respectively. Various couplings are proposed where interactions are achieved through the causal influence between state variables. We compare non-coupled and coupled models on two tasks: the recognition of artificially degraded handwritten digits and the recognition of real degraded old printed characters. Our models show that coupled architectures perform more accurately on degraded characters than basic HMMs, the linear combination of independent HMM scores, as well as discriminative methods such as support vector machines (SVMs).

This paper presents a new approach for text-line segmentation based on Block Covering which solves the problem of overlapping and multi-touching components. Block Covering is the core of a system which processes a set of ancient Arabic documents from historical archives. The system is designed for separating text-lines even if they are overlapping and multi-touching. We exploit the Block Covering technique in three steps: a new fractal analysis (Block Counting) for document classification, a statistical analysis of block heights for block classification and a neighboring analysis for building text-lines. The Block Counting fractal analysis, associated with a fuzzy C-means scheme, is performed on document images in order to classify them according to their complexity: tightly (closely) spaced documents (TSD) or widely spaced documents (WSD). An optimal Block Covering is applied on TSD documents which include overlapping and multi-touching lines. The large blocks generated by the covering are then segmented by relying on the statistical analysis of block heights. The final labeling into text-lines is based on a block neighboring analysis. Experimental results provided on images of the Tunisian Historical Archives reveal the feasibility of the Block Covering technique for segmenting ancient Arabic documents.

Conference of 13th International Conference on Document Analysis and Recognition, ICDAR 2015 ; Conference Date: 23 August 2015 Through 26 August 2015; Conference Code:118256; International audience; Handwriting recognition systems usually rely on static dictionaries and language models. Full coverage of these dictionaries is generally not achieved when dealing with unrestricted document corpora due to the presence of Out-Of-Vocabulary words. In a previous work, dynamic dictionaries were built from Web resources and successfully applied to isolated word recognition. In the present work we extend this approach to text-line recognition. Line segmentation into words is needed to exploit dynamic dictionaries and it is performed using BLSTM classifiers to align filler models and word sequence outputs. Words are then classified based on the confidence score into anchor and non-anchor words (AWs and NAWs). AWs are equated to the BLSTM outputs and used as such. Dynamic dictionaries are built for NAWs by exploiting Web resources for their character sequence and for neighboring AWs. Text-lines are decoded again using dynamic dictionaries and re-estimated language model. We conduct experiments on the publicly available RIMES database and show that the introduction of the dynamic dictionary is beneficial. Equally important, we show that the gain increases as the proportion of OOVs increases.

International audience; Handwriting recognition systems usually rely on static dictionaries and language models. Full coverage of these dictionaries is generally not achieved when dealing with unrestricted document corpora due to the presence of Out-Of-Vocabulary (OOV) words. We propose an approach which uses the World Wide Web as a corpus to improve dictionary coverage. We exploit the very large and freely available Wikipedia corpus in order to obtain dynamic dictionaries on the fly. We rely on recurrent neural network (RNN) recognizers, with and without linguistic resources, to detect words that are non-reliably recognized within a word sequence. Such words are labeled as non-anchor words (NAWs) and include OOVs and In-Vocabulary words recognized with low confidence. To recognize a non-anchor word, a dynamic dictionary is built by selecting words from the Web resource based on their string similarity with the NAW image, and their linguistic relevance in the NAW context. Similarity is evaluated by computing the edit distance between the sequence of characters generated by the RNN recognizer exploited as a filler model, and the Wikipedia words. Linguistic relevance is based on an N-gram language model estimated from the Wikipedia corpus. Experiments conducted on aword-segmented version of the publicly available RIMES database show that the proposed approach can improve recognition accuracy compared to systems based on static dictionaries only. The proposed approach shows even better behavior as the proportion of OOVs increases, in terms of both accuracy and dictionary coverage.

Humans produce handwriting samples in their every-day life activity. Handwriting can be collected on-line, thanks to digitizing tablets, or off-line by scanning paper sheets. One question is whether such handwriting data can tell about individual style, health or emotional state of the writer. We will try to answer this question by presenting related works conducted on on-line and off-line data. We will show that pieces of off-line handwriting may be useful for recognizing a writer or a handwriting style. We will also show that anxiety can be recognized as an emotion from on-line handwriting and drawings using a non parametric approach such as random forests.

Arabic handwritten documents present specific challenges due to the cursive nature of the writing and the presence of diacritical marks. Moreover, one of the largest labeled database of Arabic handwritten documents, the OpenHart-NIST database includes specific noise, namely guidelines, that has to be addressed. We propose several approaches to process these documents. First a guideline detection approach has been developed, based on K-means, that detects the documents that include guidelines. We then propose a series of preprocessing at text-line level to reduce the noise effects. For text-lines including guidelines, a guideline removal preprocessing is described and existing keystroke restoration approaches are assessed. In addition, we propose a preprocessing that combines noise removal and deskewing by removing line fragments from neighboring text lines, while searching for the principal orientation of the text-line. We provide recognition results, showing the significant improvement brought by the proposed processings.

17 pages, 11 figures.; International audience; The problem addressed in this paper is the automatic extraction of names from a document image. Our approach relies on the combination of two complementary analyses. First, the image-based analysis exploits visual clues to select the regions of interest in the document. Second, the textual-based analysis searches for name patterns and low-level word textual features. Both analyses are then combined at the word level through a neural network fusion scheme. Reported results on degraded documents such as facsimile and photocopied technical journals demonstrate the interest of the combined approach.

International audience; Handwriting recognition systems rely on predefined dictionaries obtained from training data. Small and static dictionaries are usually exploited to obtain high in-vocabulary (IV) accuracy at the expense of coverage. Thus the recognition of out-of-vocabulary (OOV) words cannot be handled efficiently. To improve OOV recognition while keeping IV dictionaries small, we introduce a multi-step approach that exploits Web resources. After an initial IV-OOV sequence classification, external resources are used to create OOV sequence-adapted dynamic dictionaries. A final Viterbi-based decoding is performed over the dynamic dictionary to determine the most probable word for the OOV sequence. We validate our approach with experiments conducted on RIMES, a publicly available database. Results show that improvements are obtained compared to standard handwriting recognition, performed with a static dictionary. Both domainadapted and generic dynamic dictionaries are studied and we show that domain adaptation is beneficial.

This paper deals with the recognition of free-style handwritten text lines. We compare 2 state-of-the-art segmentation-free recognition approaches. The first one is the popular context-dependent HMM approach (Hidden Markov Models). The second one is the recent BLSTM (Bi-directional Long Short-Term Memory) approach based on recurrent neural networks and memory blocks. For the sake of comparison, both recognizers use the same set of features and language model. They are compared from the following perspectives: sliding window parameters for feature extraction, training and decoding speed and performance accuracy with or without using a language model. We compare these two approaches on the publicly available Rimes database of French handwritten mails. Our main findings are that long frame sequences, obtained with specific window parameters, improve both recognizers, and that BLSTMs outperform HMMs in terms of WER rates, at the expense of considerably longer training times.

International audience; Handwriting recognition systems are typically trained using publicly available databases, where data have been collected in controlled conditions (image resolution, paper background, noise level, etc.). Since this is not often the case in real-world scenarios, classification performance can be affected when novel data is presented to the word recognition system. To overcome this problem, we present in this paper a new approach called database adaptation. It consists of processing one set (training or test) in order to adapt it to the other set (test or training, respectively). Specifically, two kinds of preprocessing, namely stroke thickness normalization and pixel intensity normalization are considered. The advantage of such approach is that we can re-use the existing recognition system trained on controlled data. We conduct several experiments with the Rimes 2011 word database and with a real-world database. We adapt either the test set or the training set. Results show that training set adaptation achieves better results than test set adaptation, at the cost of a second training stage on the adapted data. Accuracy of data set adaptation is increased by 2% to 3% in absolute value over no adaptation.

International audience; This paper presents a system for the recognition of unconstrained handwritten mails. The main part of this system is an HMM recognizer which uses trigraphs to model contextual information. This recognition system does not require any segmentation into words or characters and directly works at line level. To take into account linguistic information and enhance performance, a language model is introduced. This language model is based on bigrams and built from training document transcriptions only. Different experiments with various vocabulary sizes and language models have been conducted. Word Error Rate and Perplexity values are compared to show the interest of specific language models, fit to handwritten mail recognition task.

We present in this paper an HMM-based recognizer for the recognition of unconstrained Arabic handwritten words. The recognizer is a context-dependent HMM which considers variable topology and contextual information for a better modeling of writing units. We propose an algorithm to adapt the topology of each HMM to the character to be modeled. For modeling the contextual units, a state-tying process based on decision tree clustering is introduced which significantly reduces the number of parameters. Decision trees are built according to a set of expert-based questions on how characters are written. Questions are divided into global questions yielding larger clusters and precise questions yielding smaller ones. We apply this modeling to the recognition of Arabic handwritten words. Experiments conducted on the OpenHaRT2010 database show that variable length topology and contextual information significantly improves the recognition rate.

We propose a novel scenario called "writer retrieval" consisting in the retrieval from a set of documents all those produced by the same writer. The retrieval is based on a set of ten features correlated to perceived characteristics of the writing (orientation, regularity...). These features are extracted from the probability density function of the orientations of the writing contour. We evaluate the overall efficiency of each feature and the relative importance of features according to queries. Evaluation is conducted on the publicly available IAM database.

This study aims at building an efficient word recognition system resulting from the combination of three handwriting recognizers. The main component of this combined system is an HMM-based recognizer which considers dynamic and contextual information for a better modeling of writing units. For modeling the contextual units, a state-tying process based on decision tree clustering is introduced. Decision trees are built according to a set of expert-based questions on how characters are written. Questions are divided into global questions, yielding larger clusters, and precise questions, yielding smaller ones. Such clustering enables us to reduce the total number of models and Gaussians densities by 10. We then apply this modeling to the recognition of handwritten words. Experiments are conducted on three publicly available databases based on Latin or Arabic languages: Rimes, IAM, and OpenHart. The results obtained show that contextual information embedded with dynamic modeling significantly improves recognition.

This paper proposes a novel method for document enhancement. The method is based on the combination of two state-of-the-art filters through the construction of a mask. The mask is applied to a TV (Total Variation) - regularized image where background noise has been reduced. The masked image is then filtered by NLmeans (Non LocalMeans) which reduces the noise in the text areas located by the mask. The document images to be enhanced are real historical documents from several periods which include several defects in their background. These defects result from scanning, paper aging and bleed-through. We observe the improvement of this enhancement method through OCR accuracy.

This article describes the isolated word recognizer presented by the authors to the ICDAR 2009 French handwriting recognition competition. The system is a combination of three isolated word recognizers based on different features and models. A novel n-best combination method is proposed and compared to standard combination methods. New results on the ICDAR 2009 test database are reported.

This paper presents an HMM-based recognizer for the off-line recognition of handwritten words. Word models are the concatenation of context-dependent character models (trigraphs). The trigraph models we consider are similar to triphone models in speech recognition, where a character adapts its shape according to its adjacent characters. Due to the large number of possible context-dependent models to compute, a top-down clustering is applied on each state position of all models associated with a particular character. This clustering uses decision trees, based on rhetorical questions we designed. Decision trees have the advantage to model untrained trigraphs. Our system is shown to perform better than a baseline context independent system, and reaches an accuracy higher than 74% on the publicly available Rimes database.

The effects of different image pre-processing methods for document image binarization are explored. They are compared on five different binarization methods on images with bleed through and stains as well as on images with uniform background speckle. The binarization method is significant in the binarization accuracy, but the pre-processing also plays a significant role. The Total Variation method of pre-processing shows the best performance over a variety of pre-processing methods.

We compare in this study two image restoration approaches for the pre-processing of printed documents:namely the Non-local Means filter and a total variation minimization approach. We apply these two approaches to printed document sets from various periods,and we evaluate their effectiveness through character recognition performance using an open source OCR. Our results show that for each document set, one or both pre-processing methods improve character recog-nition accuracy over recognition without preprocessing. Higher accuracies are obtained with Non-local Means when characters have a low level of degradation since they can be restored by similar neighboring parts of non-degraded characters. The Total Variation approach is more effective when characters are highly degraded and can only be restored through modeling instead of using neighboring data.

This paper describes a knowledge based system that helps scribes to authenticate Hebrew manuscripts for which accurate laws of calligraphy have been given to the scribes. Paleographic expertise is also included in order to characterize the size of the document and the writing. When used for authentication purposes, the system shortens the task of the scribe by pointing out the parts of the document or the characters where problems arise for the machine (low contrast, breaks, ambiguous shape). The scribe will then operate on a restricted part of the document and decide whether it has to be corrected or not. In order to be close to human mechanisms of interpretation, the system structure includes several analysis levels. This system uses some principles of intelligent vision systems that can set up image and document analysis strategies as well as interactions between the high level and low level procedures. It works directly on grey level pictures.

In this paper we present a system for the off-line recognition of cursive Arabic handwritten words. This system in an enhanced version of our reference system presented in [El-Hajj et al., 05] which is based on Hidden Markov Models (HMMs) and uses a sliding window approach. The enhanced version proposed here uses contextual character models. This approach is motivated by the fact that the set of Arabic characters includes a lot of ascending and descending strokes which overlap with one or two neighboring characters. Additional character models are constructed according to characters in their left or right neighborhood. Our experiments on images of the benchmark IFN/ENIT database of handwritten villages/towns names show that using contextual character models improves recognition. For a lexicon of 306 name classes, accuracy is increased by 0.6% in absolute value which corresponds to a 7.8% reduction in error rate.

This paper presents a text line segmentation method for printed or handwritten historical Arabic documents. Documents are first classified into 2 classes using a K-means scheme. These classes correspond to document complexity (easy or not easy to segment). Then, a document which includes overlapping and touching characters, is divided into vertical strips. The extracted text blocks obtained by horizontal projection are classified into three categories: small, average and large text blocks. After segmenting the large text blocks, the lines are obtained by matching adjacent blocks within two successive strips using spatial relationship. The document without overlapping or touching characters is segmented by making abstraction on the segmentation module of the large text blocks. The text line segmentation method has a 96% accuracy on a collection of 100 historical documents

This paper investigates the application of dynamic Bayesian networks (DBNs) to the recognition of degraded characters from historical printed books. This framework allows us to capture the 2D nature of character images by the coupling of two HMMs (Hidden Markov Models). The vertical HMM observes image columns while the horizontal HMM observes image rows respectively. Two coupled DBN architectures are proposed to model interactions between these two streams. We present experiments on real degraded characters extracted from an ancient printed book (17th century). These experiments demonstrate that coupled architectures significantly better cope with broken characters than non coupled ones and than discriminative methods such as SVMs.

We investigate in this paper the application of dynamic Bayesian networks (DBNs) to the recognition of handwritten digits. The main idea is to couple two separate HMMs into various architectures. First, a vertical HMM and a horizontal HMM are built observing the evolving streams of image columns and image rows respectively. Then, two coupled architectures are proposed to model interactions between these two streams and to capture the 2D nature of character images. Experiments performed on the MNIST handwritten digit database show that coupled architectures yield better recognition performances than non-coupled ones. Additional experiments conducted on artificially degraded (broken) characters demonstrate that coupled architectures better cope with such degradation than non coupled ones and than discriminative methods such as SVMs.

In this paper, we describe a 1D HMM offline handwriting recognition system employing an analytical approach. The system is supported by a set of robust language independent features extracted on binary images. Parameters such as lower and upper baselines are used to derive a subset of baseline dependent features. Thus, word variability due to lower and upper parts of words is better taken into account. In addition, the proposed system learns character models without character pre-segmentation. Experiments that have been conducted on the benchmark IFN/ENIT database of Tunisian handwritten country/village names, show the advantage of the proposed approach and of the baseline-dependant features.

5 pages, 2 figures, composed with Microsoft Word. Proceedings of the Seventh International Conference on Document Analysis and Recognition (ICDAR 2003). Edinburgh, Écosse, 3-6 août 2003. Également disponible à l'URL : http://www.cse.salford.ac.uk/prima/ICDAR2003/Papers/0100_563_likforman_l.pdf; International audience; In the frame of a Unified Messaging System, a crucial task of the system is to provide the user with key information on every message received, like keywords reflecting the object of the message, or the name of the sender. However, in the case of facsimiles, this information is not as easy to detect as in the case of e-mails, since no standard headers are defined. The aim of the present work is to identify and extract a specific information (the name of the sender) from a fax cover page. For this purpose, methods based on image document analysis (OCR recognition, physical blocks selection), and text analysis methods (optimised dictionary lookup, local grammar rules), are implemented to work in parallel. The fusion of their results brings a more accurate guess than any of the methods would achieve separately.

A comparison is made between several hidden Markov models in the context of printed character recognition. Two HMMs are first compared, one dealing with columns of a character image, and the other dealing with lines. These 2 HMMs are then associated in a decision fusion scheme combining the log-likelihoods provided by each HMM classifier. The statistical assumptions underlying the combination formula are described and the combination formula is shown to be an approximation of a real joint log-likelihood. The last experiment consists of building a single HMM, modeling the joint flow of lines and columns. This data fusion scheme is shown to be more accurate as it highlights correlations between the line and column features.

Presents the first achievements of HERS (Hypermedia Edit and Read Station), which is devoted to the browsing and editing of hypermedia documents from literary material including document images. Concerning the editing, our purpose as twofold. First, capabilities are offered to transcribe manuscripts. Transcribing the text consists of coupling lines typed on the keyboard with their corresponding text lines in the manuscript images. A collaborative system, based on computer-human interaction and document analysis, is proposed for performing this task. Second, interactive tools are offered to organize the electronic document and establish hypermedia links between its different components (image areas, transcribed words or lines, or other kinds of heterogeneous data). Concerning the browsing, we developed an approach based on information visualization in order to provide users with an idea of the overall organization of the hyperdocument and so help them to navigate through it.

The method herein proposed detects text lines on handwritten pages which may include either lines oriented in several directions, erasures, or annotations between main lines. The method has a hypothesis-validation strategy which is iteratively activated until the end of the segmentation is reached. At each stage of the process, the best text-line hypothesis is generated in the Hough domain. Taking into account the fluctuations of the text-line components. Afterwards, the validity of the line is checked in the image domain using a proximity criteria which analyses the context in which is perceived the alignment hypothesized. Ambiguous components belonging to several text lines are also marked.

We discuss the name block location process of a messaging system which deals with facsimile images including handwritten items. This process is the first stage of an information extraction task which aims at extracting condensed information from such documents. The approach is based on the use of spatial/visual cues for selecting regions in the image where sender and recipient information can be found. These cues are based on the human visual grouping process. The start of the process is the binary image and its electronic version produced by an OCR system. A document processing stage first extracts layout components at a pseudo word level and detects handwritten components. The electronic version enables the selection of layout components including predefined printed keywords. These layout components are then grouped under spatial constraints to initialize a logical description. The logical objects searched for are the main field headers relative to the sender and the recipient and the sender name. Experimental results on the performance of the approach are presented.

Many preprocessing techniques have been proposed for isolated word recognition. However, recently, recognition systems have dealt with text blocks and their compound text lines. In this paper, we propose a new preprocessing approach to efficiently correct baseline skew and fluctuations. Our approach is based on a sliding window within which the vertical position of the baseline is estimated. Segmentation of text lines into subparts is, thus, avoided. Experiments conducted on a large publicly available database (Rimes), with a BLSTM (bidirectional long short-term memory) recurrent neural network recognition system, show that our baseline correction approach highly improves performance.

This paper presents a novel approach for the multi-oriented text line extraction from historical handwrittenArabic documents. Because of the multi-orientation of lines and their dispersion in the page, we use an imagepaving algorithm that can progressively and locally determine the lines. The paving algorithm is initialized witha small window and then its size is corrected by extension until enough lines and connected components werefound. We use the Snake for line extraction. Once the paving is established, the orientation is determined usingthe Wigner-Ville distribution on the histogram projection pro“le. This local orientation is then enlarged to limitthe orientation in the neighborhood. Afterwards, the text lines are extracted locally in each zone basing onthe follow-up of the baselines and the proximity of connected components. Finally, the connected componentsthat overlap and touch in adjacent lines are separated. The morphology analysis of the terminal letters ofArabic words is here considered. The proposed approach has been experimented on 100 documents reaching anseparation accuracy of about 98.6%.Keywords: Handwritten Arabic documents, text line segmentation, skew angle estimation, Snake, Wigner-Villedistribution, overlapping and touching lines.