This paper proposes a fully automatic new method for generating semi-synthetic images of historical documents to increase the number of training samples in small datasets. This method extracts and mixes background only images (BOI) with text only images (TOI) issued from two different sources to create semi-synthetic images. The TOIs are extracted with the help of a binary mask obtained by binarizing the image. The BOIs are reconstructed from the original image by replacing TOI pixels using an inpainting method. Finally, a TOI can be efficiently integrated in a BOI using the gradient domain, thus creating a new semi-synthetic image. The idea behind this technique is to automatically obtain documents close to real ones with different backgrounds to highlight the content. Experiments are conducted on the public HisDB dataset which contains few labeled images. We show that the proposed method improves the performance results of a semantic segmentation and baseline extraction task.

In this work, we investigate the use of the attention mechanism in deep learning for a better reading of handwritten digit strings in digitized images. The proposed recognition system built upon a CNN (Convolutional Neural Network) and two RNNs (Recurrent Neural Networks), acting as Encoder and Decoder and using the attention mechanism. We used a 1D mechanism for attention location with a "soft" alignment attention which has the peculiarity of having an easily calculable gradient and thus to integrate well with the network. Experimental results on data from ORAND-CAR A, ORAND-CAR B and CVL HDS databases compare favorably to other published methods.

This paper proposes a new evaluation metric based on the existing ZoneMap metric. The ZoneMap method, designed to perform a zone segmentation evaluation and classification, is considered in the context of OCR evaluation. Its limits are spotted, described and a new algorithm, ZoneMapAlt (ZoneMap Alternative) is proposed to solve the identified limits while keeping the properties of the original one. To validate the new metric, experiments have been made on a dataset of scientific articles. Results demonstrate that the ZoneMapAlt algorithm provides greater details on segmentation errors and is able to detect critical segmentation errors.

This paper presents a complete procedure that uses contextual and syntactic information to identify and recognize amount fields in the table regions of chemistry documents. The proposed method is composed of two main modules. Firstly, a structural analysis based on connected component (CC) dimensions and positions identifies some special symbols and clusters other CCs into three groups: fragment of characters, isolated characters or connected characters. Then, a specific processing is performed on each group of CCs. The fragment of characters are merged with the nearest character or string using geometric relationship based rules. The characters are sent to a recognition module to identify the numeral components. For the connected characters, the final decision on the string nature (numeric or non-numeric) is made based on a global score computed on the full string using the height regularity property and the recognition probabilities of its segmented fragments. Finally, a simple syntactic verification at table row level is conducted in order to correct eventual errors. The experimental tests are carried out on real-world chemistry documents provided by our industrial partner eNovalys. The obtained results show the effectiveness of the proposed system in extracting amount fields.

This paper extends a previous work being done by [4]. Having no information about the document separation in the flow, the system operates progressively by examining successive pairs of pages looking for continuity or rupture descriptors. Four document levels have been introduced to better extract those descriptors and reduce the ambiguity in their extraction: records, technical documents, fundamental documents and cases. At each level, structural and factual descriptors are first extracted and then compared between pairs of pages or documents. To reinforce the descriptor interest and focus the system on equivalent descriptors in the pairs, the descriptors are accompanied by their context. The extraction of the context is facilitated by the determination of the physical and logical structure in the pages. Contextual rules based on these descriptors are employed for the determination of either a continuity, a rupture or an uncertainty between the pairs. To overcome the problem of information emptiness in the current page, a logbook is used to gather the descriptors in all the previous pages of the record and a buffer allows to delay the comparison. These latter points were added to the previous work that widely reinforce the current system increasing its precision of more than 6%.

This paper proposes an entity local structure comparison approach based on inexact subgraph matching. The comparison results are used for mislabeling correction in the local structure. The latter represents a set of entity attribute labels which are physically close in a document image. It is modeled by an attributed graph describing the content and presentation features of the labels by the nodes and the geometrical features by the arcs. A local structure graph is matched with a structure model which represents a set of local structure model graphs. The structure model is initially built using a set of well chosen local structures based on a graph clustering algorithm and is then incrementally updated. The subgraph matching adopts a specific cost function that integrates the feature dissimilarities. The matched model graph is used to extract the missed labels, prune the extraneous ones and correct the erroneous label fields in the local structure. The evaluation of the structure comparison approach on 525 local structures extracted from 200 business documents achieves about 90% for recall and 95% for precision. The mislabeling correction rates in these local structures vary between 73% and 100%.

This work aims to segment touching components (TCs) which may occur between word letters of consecutive text-lines or those of words of the same line in Arabic manuscripts. The proposed approach is mainly based on two steps: 1) finding for a localized touching component its most similar model, stored in a dictionary with its correct segmentation, based on shape context descriptor, 2) segmenting the touching component based on central point of the found most similar model's parts. Tests are performed using a database of connection zones (1300 samples) and three metrics: Manhattan, Euclidean and Canberra distances. Experimental results have shown the effectiveness of the proposed touching component segmentation method in comparison to some related works. Our best achieved TC segmentation rate is of 94%.

In this paper we addressed the problem of Arabic mathematical formula recognition, extracted from scanned images of clearly printed documents. Two main stages are followed by the proposed system: symbol recognition and structural analysis of the mathematical formula. For the first stage, our system uses a combination of different statistical features like Run length, Hu and Zernike moments, Bi-level co-occurrence and white pixel's portion and an instance-based classifier K*. High accuracy for the recognition of isolated mathematical symbols is achieved. In the second stage, the system proceeds by top-down and bottom-up parsing scheme based on operator dominance. A set of replacement rules is defined by a coordinate grammar based on symbol recognition and symbol arrangement analysis results. In the proposed system, the recognition and parsing modules interact more closely. Thus, we can use the context information collected during structural analysis to help us guess about the symbols, overcoming our incorrect assumption of perfect symbol recognition. The system provides output in MathML which is easily transmitted for subsequent processing by computer algebra systems. The syntax-directed recognition system, described here, has been successfully demonstrated in many types of formulas and achieved satisfactory results. 91% of formulas are correctly recognized.

In this paper we present a separator line and content analysis based approach for table structure extraction in handwritten chemistry documents. A first module based on Hough Transform technique is used to detect all graphic lines in a document. The resulting grid is analyzed in order to find the cell boundaries. In case of absence of these lines, a second module uses content information to define boundaries between cells. The digits, representing the dominant components in the handled tables, are identified using a multistage classification system. Then, the digit cartography is analyzed based on syntactical rules in order to find cell boundaries. The proposed method has been tested on a set of handwritten chemistry documents and experimental results indicate satisfactory performance.

In this paper, we present a query-based approach to selectively extract tabular information and recognize the table structure from scanned documents. Unlike conventional table processing paradigms, we adopt a client-driven approach where clients provide a query pattern by specifying a set of key-fields in the document image. The query pattern is first transformed into an attributed relational graph where each node is described with features and the edges with spatial relationships between the nodes. A fast graph matching technique is then used to retrieve other similar graphs from the document image. Further, the extracted graphs are collectively analyzed to deduce the overall tabular structure. Experiments on a dataset of 101 commercial transaction documents demonstrate the effectiveness of the proposed method.

In this work, we investigate the combination of PGM (Propabilistic Graphical Models) classifiers, either independent or coupled, for the recognition of Arabic handwritten words. The independent classifiers are vertical and horizontal HMMs (Hidden Markov Models) whose observable outputs are features extracted from the image columns and the image rows respectively. The coupled classifiers associate the vertical and horizontal observation streams into a single DBN (Dynamic Bayesian Network). A novel method to extract word baseline and a simple and easily extractable features to construct feature vectors for words in the vocabulary are proposed. Some of these features are statistical, based on pixel distributions and local pixel configurations. Others are structural, based on the presence of ascenders, descenders, loops and diacritic points. Experiments on handwritten Arabic words from IFN/ENIT strongly support the feasibility of the proposed approach. The recognition rates achieve 90.42% with vertical and horizontal HMM, 85.03% and 85.21% with respectively a first and a second DBN which outperform results of some works based on PGMs.

This paper proposes an entity recognition approach in scanned documents referring to their description in database records. First, using the database record values, the corresponding document fields are labeled. Second, entities are identified by their labels and ranked using a TF/IDF based score. For each entity, local labels are grouped into a graph. This graph is matched with a graph model (structure model) which represents geometric structures of local entity labels using a specific cost function. This model is trained on a set of well chosen entities semi-automatically annotated. At the end, a correction step allows us to complete the eventual entity mislabeling using geometrical relationships between labels. The evaluation on 200 business documents containing 500 entities reaches about 93% for recall and 97% for precision.

In this paper, we propose a new scheme for script and nature identification. The objective is to discriminate between machine-printed/handwritten and Latin/Arabic scripts at word level. It is relatively a complex task due to possible use of multi-fonts and sizes, complexity and variation in handwriting. In the proposed script identification system, we extract features from word images using Co-occurrence Matrix of Oriented Gradients (Co-MOG). The classification is done using different classifiers. Extensive experimentation has been carried on 24000 words, extracted from standard databases. An average identification accuracy of 99.85% is achieved by k Nearest Neighbors (k-NN) classifier which clearly outperforms results of some existing systems.

We consider an industrial context where we deal with a stream of unlabelled documents that become available progressively over time. Based on an adaptive incremental neural gas algorithm (AING), we propose a new stream-based semi supervised active learning method (A2ING) for document classification, which is able to actively query (from a human annotator) the class-labels of documents that are most informative for learning, according to an uncertainty measure. The method maintains a model as a dynamically evolving graph topology of labelled document-representatives that we call neurons. Experiments on different real datasets show that the proposed method requires on average only 36.3% of the incoming documents to be labelled, in order to learn a model which achieves an average gain of 2.15-3.22% in precision, compared to the traditional supervised learning with fully labelled training documents.

Our ultimate objective is to contribute to the field of script and nature identification to be able to differentiate, at word level, handwritten or machine-printed, Arabic and Latin scripts. Different sets of features have been employed successfully for discriminating between Arabic and Latin words. They include few well-established features previously used and adapted in our case and new structural features which are intrinsic features of Arabic and Latin scripts. We select features that maximize the distinction between Arabic and Latin words. Experiments have been conducted with 1320 handwritten and printed words, covering a wide range of fonts, and encouraging results have been obtained. We achieved a correct classification of 98.4 percent for word level script and nature identification using Bayes classifier.

In this paper, we present a model-based document information content extraction approach and perform in-depth evaluation based on clients' relevance. Real-world users i.e., clients first provide a set of key fields from the document image which they think are important. These are used to represent a graph where nodes (i.e., fields) are labelled with dynamic semantics including other features and edges are attributed with spatial relations. Such an attributed relational graph (ARG) is then used to mine similar graphs from a document image that are used to reinforce or update the initial graph iteratively each time we extract them, in order to produce a model. Models therefore, can be employed in the absence of clients. We have validated the concept and evaluated its scientific impact on real-world industrial problem, where table extraction is found to be the best suited application.

This paper addresses the use of PGM (Probabilistic Graphical Model) for form model identification from just few items filled up by an electronic pen. Only the electronic ink is sent to the system without any indication on the form model. Two applications are made in this study: one is related to keynote form classification from its filled fields, while the second application concerns a design modelling problem for the on-line configuration of shower areas. In the former, only indications on the filled fields are sent to the system, while in the latter, the designer send strokes corresponding to the elements designed on the form model. In this application a unique form is proposed to the user to fill up the configuration of his shower area. The PGM is exploited advantageously in both cases translating precisely the relationships between corresponding elements in conditional probabilities, from individual elements up to the complete model constitution.

In our previous research, we worked on on-line form recognition by exploiting semantic constraints between fields using Bayesian networks. The semantic constraints allowed us to check the co-existence of fields filled up by hand by users. In this paper, we propose to test the use of architectural constraints for a design problem related to the modelling of shower areas. The proposed method exploits the physical dependencies between different parts of a space shower. The tests are performed on a database composed of 500 forms representing 5 models. The first results reach a recognition rate of 96.7%.

In this paper we present a student information sheet reading system. Relevant algorithm is proposed to locate and label handwritten answer field. As information sheets can be filled in Arabic and/or in French, automating the script language differentiation is a pre-recognition required in the proposed system. We have developed a robust and fast field classification and script language identification method, based on a decision tree, to make these processing practical for sheet recognition. To this end, the system uses several novel features (loops, descenders, diacritics) and analyses the lower profile of script. The classification rates are 92.5% for numeric fields, 94.34% for Arabic scripts and 94.66% for French scripts. Experimental results, carried on 80 sheets, show our system provides an effective way to convert printed sheets into computerized format or collect information for database from printed sheets.

One of the main challenges in off-line signature verification systems is to make them robust against rotation of the signatures. A new technique for rotation invariant feature extraction based on a circular grid is proposed in this paper. Graphometric features for the circular grid are defined by adapting similar features available for rectangular grids, and the property of rotation invariance of the Discrete Fourier Transform (DFT) is used in order to achieve robustness against rotation. A Support Vector Machine (SVM) based classifier scheme is used for classification tasks. Experimental results on a public database show that the proposed verification system has a performance comparable to similar state-of-the-art signature verification systems with the additional advantage of being robust against rotation of the signatures.

This paper presents an end-to-end administrative document analysis system. This system uses case-based reasoning in order to process documents from known and unknown classes. For each document, the system retrieves the nearest processing experience in order to analyze and interpret the current document. When a complete analysis is done, this document needs to be added to the document database. This requires an incremental learning process in order to take into account every new information, without losing the previous learnt ones. For this purpose, we proposed an improved version of an already existing neural network called Incremental Growing Neural Gas. Applied on documents learning and classification, this neural network reaches a recognition rate of 97.63%.

In this paper, we present a novel approach for the multi-oriented text line extraction from handwritten Arabic documents. After image pre-processing, the local orientations are determined in small windows obtained by image paving. The orientation of the text within each window is estimated using the projection profile technique considering several projection angles. Then, the windows which close angles are gathered into largest zones. We use the Wigner-Ville Distribution (WVD) to estimate the global orientation of each zone. The WVD is more precise than the classical projection profile technique. Afterwards, the text lines are extracted in each zone basing on the follow-up of the baselines and the proximity of connected components. The experimental results prove the efficiency of the proposed scheme. It has been evaluated on 50 documents reaching an accuracy of about 97.6%.

This paper presents the XML-based formats ALTO, TEI, METS used for Digital Libraries and their interest for data representation in a Document Image Analysis and Recog- nition (DIAR) process. In the first part we briefly present these formats with focus on their adequacy for structural representation and modeling of DIAR data. The second part shows how these formats can be used in a reverse engineer- ing process. Their implementation as a data representation framework will be shown.

This paper shows the use of case-based reasoning (CBR) for invoice structure extraction and analysis. This method, called CBR-DIA (CBR for document invoice analysis), is adaptive and does not need any previous training. It analyses a document by retrieving and analysing similar documents or elements of documents (cases) stored in a database. The retrieval step is performed thanks to graph comparison techniques like graph probing and edit distance. The analysis step is done thanks to the information found in the nearest retrieved cases. Applied on 950 invoices, CBR-DIA reaches a recognition rate of 85.29% for documents of known classes and 76.33% for documents of unknown classes.

In this paper, we propose a rejection strategy for convolutional neural network models. The purpose of this work is to adapt the network's topology injunction of the geometrical error. A self-organizing map is used to change the links between the layers leading to a geometric image transformation occurring directly inside the network. Instead of learning all the possible deformation of a pattern, ambiguous patterns are rejected and the network's topology is modified in function of their geometric errors thanks to a specialized self-organizing map. Our objective is to show how an adaptive topology, without a new learning, can improve the recognition of rejected patterns in the case of handwritten digits.

In spite of the improvement of commercial optical character recognition (OCR) during the last years, their ability to process different kinds of documents can also be a default. They cannot produce a perfect recognition for all documents. However they allow producing high result for standard cases. We propose in this paper a model combining several OCRs and a specialized ICR (intelligent character recognition) based on a convolutional neural network to complement them. Instead of just performing several OCRs in parallel and applying a fusing rule of the results, a specialized neural network with an adaptive topology is added to complement the OCRs in function of the OCRs errors. This system has been tested on ancient documents containing old characters and old fonts not used in contemporary documents. The OCRs combination increases the recognition of about 3% whereas the ICR improves the recognition of rejected characters of more than 5%.

This paper introduces three neural based binarization techniques. These techniques start with a self organizing map (SOM) applied on the image to extract its most representative grey levels or colors. The classification goes further in two different ways. In the case of grey level images, the Kmeans algorithm or Sauvola's or Niblack's thresholds are used, whereas a multi layer perceptron (MLP) is used in the case of color images. The obtained results are discussed and we show that they are better than those of some classical binarization techniques.

The purpose of this work is to develop a pattern recognition system simulating the human vision. A transparent neural network, with context returns is used. The context returns consist in using global vision to correct local vision (i.e. input data are corrected according to neural network outputs). In order not to compute all the input features during these context returns, a filter-based method was designed to organize the features in clusters. This allows finding a good subset of input features during each cycle, which reduce the computations. The method interest is shown in the case of logical document structure retrieval.

In this paper, we present a system towards Indian postal automation based on the recognition of pin-code and city name of the postal document. In the proposed system, at first, non-text blocks (postal stamp, postal seal etc.) are detected and destination address block (DAB) is identified from the document. Next, lines and words of the DAB are segmented. Since India is a multi-lingual and multi-script country, the address part may be written by combination of two scripts. To identify the script by which a word is written, we propose a water reservoir based technique. It is very difficult to identify the script by which the pin-code portion is written. So, we have used two-stage artificial neural network (NN) based general classifiers for the recognition of pin-code digits written in English/Bangla. For recognition of city names, we propose an NSHP-HMM (non-symmetric half plane-hidden Markov model) based technique.

In this paper, an improvement of a 2D stochastic model based handwritten entity recognition system is described. To model the handwriting considered as being a two dimensional signal, a context based, segmentation-free hidden Markov model (HMM) recognition system was used. The baseline approach combines a Markov random field (MRF) and a HMM so-called non-symmetric half plane hidden Markov model (NSHP-HMM). To improve the results performed by this baseline system operating just on low-level pixel information an extension of the NSHP-HMM is proposed. The mechanism allows to extend the observations of the NSHP-HMM by implanting structural information in the system. At present, the accuracy of the system on the SRTP1 French postal check database is 87.52% while for the handwritten Bangla city names is 86.80%. The gain using this structural information for the SRTP dataset is 1.57%.

In this paper, a method based on part-of-speech tagging (PoS) is used for bibliographic reference structure. This method operates on a roughly structured ASCII file, produced by OCR. Because of the heterogeneity of the reference structure, the method acts in a bottom-up way, without an a priori model, gathering structural elements from basic tags to sub-fields and fields. Significant tags are first grouped in homogeneous classes according to their grammar categories and then reduced in canonical forms corresponding to record fields: "authors", "title", "conference name", "date", etc. Non labelled tokens are integrated in one or another field by either applying PoS correction rules or using a structure model generated from well-detected records. The designed prototype operates with a great satisfaction on different record layouts and character recognition qualities. Without manual intervention, 96.6% words are correctly attributed, and about 75.9% references are completely segmented from 2500 references.

In this paper, an idea for the combination of global andlocal view models is presented. These two type of modelshave proved their capabilities independantly. Some combinationwere proposed, using global view models for localanalysis, and local view models to synthetize local results.An opposite approach is proposed here : local view modelsare used as a normalization tool, while global view modelsare used for the recognition of the normalized image.The use of local view models for normalization is justifiedby their capability to perform a non-linear normalizationaccording to the image information. We propose Markovmodels as local view models, and Neural Netwok as globalview models. Using Markov models for the normalizationincreases results up to 3% better than a classical linearnormalization. Global results are improved, performing2% better than the Markov model itself. The extension ofthe system to an analytic approach is discussed.

In this paper, an approach for adaptive region segmentation of mail-order forms for high volume application is described. Regions are first identified through a selection of their anchor points described by a constraint graph, illustrating their typographic aspects in the nodes, and their topographical relationships in the arcs. Then the identification of the actual anchor points is performed from a list of textual candidates, using the Arc Consistency Algorithm (AC4). Finally, some contextual heuristics are investigated for properly delimiting the regions. The originality of this approach lies mainly in the absence of a rigid a priori model, replaced by a simple and reliable association of anchor points. The constraint graph used for their description can be easily derived from a general logical definition of their content. Experimental results are overall encouraging and the methodology integration is under execution for commercialization.

Deals with a scheme for automatic segmentation of unconstrained handwritten connected numerals. The scheme is mainly based on features obtained from a new concept based on a water reservoir. A reservoir is a metaphor to illustrate the region where numerals touch. The reservoir is obtained by considering accumulation of water poured from the top or from the bottom of the numerals. At first, considering the reservoir location and size, touching positions (top, middle and bottom) are decided. Next, by analyzing the reservoir boundary, touching position and topological features of the touching pattern, the best cutting point is determined. Finally, combined with morphological structural features the cutting path for segmentation is generated.

A major problem in handwriting recognition is the huge variability and distortions of patterns. Elastic models based on local observations and dynamic programming such HMM are efficient to absorb this variability, but their vision is local. Furthermore, global models such neural network having a fixed input size are efficient to make correlations on an entire pattern, but they cannot deal with length variability and are very sensitive to distortions. This paper proposes to use the power of these two classes of models. The elastic model is used to normalize the input-image and the fixed model performs the recognition.. The elastic model-uses an NSHP-HMM and the global model uses a support vector machine (SVM). The NSHP-HMM searches the important features and absorbs the distortions. According to the localisation of these features a pattern can be normalized to a standard size. Then the SVM is used to estimate global correlations and classify the pattern. The first results obtained are encouraging and confirm the validity of our approach.

We present a method for automatic extraction of mathematical formulas from images of documents without character recognition. Formula extraction is first done by location of its most significant symbols, then extension to adjoining symbols using contextual rules until delimitation of the whole formula space. Mathematical symbol labelling is realised from models created at the learning stage using fuzzy logic. From the experiments, we found that the average rate of primary labelling of mathematical symbols is about 95.3%. The obtained results have demonstrated the applicability of our system since 90% of mathematical formulas are well extracted from documents printed with high quality.

Presents an approach for the logical structure recognition of bibliographic references. The objective is to produce, for each reference (given in a display format such as Postscript), structured data containing the hierarchy of fields recognized. As a result of variation among bibliographic references (in the order and typographic format of fields, or writing style of the author, for example), we need a robust and tolerant system architecture. Thus, recognition is performed by a concept-oriented system that uses a model which is automatically built from a reference database. This model represents the reference fields and includes statistics on the occurrence of their terms. Recognition is achieved by a step-by-step activation of the more pertinent concepts. Each activated concept causes the execution of an appropriate searching agent. This architecture is robust and non-deterministic, allowing a solution even in difficult cases.

The use of generic model for a document class as the knowledge base in a Document Analysis System facilitates the analysis and understanding of documents belonging to this class. Nevertheless, absence of tools permitting the acquisition of this type of model is an hindrance to the conception of entirely automatic systems. In this paper, we present a method for acquiring the generic model for a document class from document samples belonging to this class. Our method is based on Inference of Tree Grammars and combination of ODA-like generic constructors. The method constructs specific physical structure for each sample and invites the user to assign logical labels to its components. From these logically labeled specific structures, it generates and modifies the generic model for the class under treatment.

We present an AI approach for the semantic recognition of bibliography references. The objective is to produce for each reference (given by an OCR flow), a structured data containing the list of the different sub-fields recognized and semantically validated. The validation is operated according to a Bibliography reference database, by the exam of principal terms in each reference field. The system uses an emergent architecture containing a Concept Network built from the database. This net represents the principal fields of the references and includes statistics on the occurrence of their terms. Validation is achieved dynamically by activation at each time of the more pertinent concepts. These concepts verify the presence of their terms by the execution of appropriate agents. This architecture is robust and non-deterministic allowing to find a solution in spite of OCR errors.

In this paper we describe an off-line handwritten word recognition system applied to the identification of literal french check amounts. It consists of three successive levels denoted as character, word and phrase level, each of them being related to the previous ones via conditional probability distributions. Training is done on character samples extracted from amount images which are modeled as trajectories in some feature space. At word level, guided by a dictionary, an internal character segmentation algorithm is used in order to maximize a global word probability measure. A stochastic grammar for a priori grammar generation probability of a phrase is proposed at the last level. Results obtained on a 1779 amounts data base provided by the SRTP are encouraging, showing our system open to further improvements.

A page segmentation method that allows one to cut a document page image into polygonal blocks as well as into classical rectangular blocks is described. The intercolumn and interparagraph gaps are extracted as horizontal and vertical lines. The points of intersection between these lines are treated as vertices of polygonal blocks. With the aid of the 4-connected chain codes and an intersection table, simple isothetic polygonal blocks are constructed from these points of intersection.>

A block image labeling method is presented. It does not assume that the blocks to be treated are already segmented or that they contain homogeneous data. It is based on connected component analysis to label the blocks' contents as small letter text, medium letter text, large letter text, graphics or photographs, giving the percentage of each of these components with respect to the surface area it occupies. It uses a recursive algorithm that allows one to improve on the result of segmentation. The performance of the method is given.>