Patrick Franco

Accented Handwritten Character Recognition Using SVM - Application to French

This paper deals with the problem of recognizing accented and non-accented characters in French handwriting. Accented characters increase the number of classes to be recognized. The performances of powerful classifier such as SVM are declined by the presence of accents. In this paper, an accented character is segmented into two parts: the root character or letter and the accent. These two parts are recognized separately, and the results are combined to rebuild the accented character. This approach avoids the combination of characters and accents that causes an increase in the number of classes to be considered. For handwritten character recognition, the combination of on-line and off-line features is used. The paper illustrates that French accented and non-accented characters and digits can be described by a combination of this kind of data. Moreover, the number of features of the combination is not necessarily very high. The experimental investigations show that the handwritten character recognition built on 45 selected features can compete with recognition rate and response time of other well known tested on standard databases such as UNIPEN and IRONOFF.

A System Based on Intrinsic Features for Fraudulent Document Detection

Paper documents still represent a large amount of information supports used nowadays and may contain critical data. Even though official documents are secured with techniques such as printed patterns or artwork, paper documents suffer from a lack of security. However, the high availability of cheap scanning and printing hardware allows non-experts to easily create fake documents. As the use of a watermarking system added during the document production step is hardly possible, solutions have to be proposed to distinguish a genuine document from a forged one. In this paper, we present an automatic forgery detection method based on documents intrinsic features at character level. This method is based on the one hand on outlier character detection in a discriminant feature space and on the other hand on the detection of strictly similar characters. Therefore, a feature set is computed for all characters. Then, based on a distance between characters of the same class, the character is classified as a genuine one or a fake one.

A Topological Measure for Image Object Recognition

All the effective object recognition systems are based on a powerful shape descriptor. We propose a new method for extracting the topological feature of an object. By connecting all the pixels constituting the object under the constraint to define the shortest path (minimum spanning tree) we capture the shape topology. The tree length is in the first approximation the key of our object recognition system. This measure (with some adjustments) make it possible to detect the object target in several geometrical configurations (translation / rotation) and it seems to have many desirable properties such as discrimination power and robustness to noise, that is the conclusion of the preliminary tests on characters and symbols.

Form recognition from ink strokes on tablet

This paper proposes a method for form recognition from handwritten input captured as digital ink on a tablet. Form recognition is an important step of form processing to read the data on a filled form. This type of recognition is different from traditional image matching (searching and retrieving) because the query image (data ink) has but few common characteristics with the retrieval images (i.e. form templates). The fact is that form structure is free in variation. It is not rare that two forms are very close in their structures and in semantic of fields. Filling in a form is also free in variation. The same form may be filled with different contents and in different ways of online context. The main idea for matching between ink strokes and form template in this paper is featureless, based on Bhattacharyya measure. The distance between the distribution of ink strokes and the distribution of form fields is the matching measure. These distributions are spatial information which is based on the crossing of coordinates of ink points and the crossing of fields to be filled. However, these coordinates are not taken on the same coordinate system. Ink point coordinates are based on the tablet coordinate system (differs from A4 format) while field coordinates are in paper size (for example, A4 format). In order to deal with this problem, affine transform is used to standardize the coordinate system. The coordinate system on the tablet is transformed into paper format system. The proposed method has been tested successfully with a high recognition rate on a set of 30 form templates. It is actually implemented on a real world application, in collaboration with an industrial partner who is specialized in software tablet solutions. The experience learned from the application also illustrates that the performance of matching method is at a high level of satisfaction in the real world.

A Conditional Random Field model for font forgery detection

Nowadays, document forgery is becoming a real issue. A large amount of documents that contain critical information as payment slips, invoices or contracts, are constantly subject to fraudster manipulation because of the lack of security regarding this kind of document. Previously, a system to detect fraudulent documents based on its intrinsic features has been presented. It was especially designed to retrieve copy-move forgery and imperfection due to fraudster manipulation. However, when a set of characters is not present in the original document, copy-move forgery is not feasible. Hence, the fraudster will use a text toolbox to add or modify information in the document by imitating the font or he will cut and paste characters from another document where the font properties are similar. This often results in font type errors. Thus, a clue to detect document forgery consists of finding characters, words or sentences in a document with font properties different from their surroundings. To this end, we present in this paper an automatic forgery detection method based on document font features. Using the Conditional Random Field a measurement of probability that a character belongs to a specific font is made by comparing the character font features to a knowledge database. Then, the character is classified as a genuine or a fake one by comparing its probability to belong to a certain font type with those of the neighboring characters.

Space-Filling Curve for Image Dynamical Indexing

In image retrieval, high-dimensional features lead often to good results, however, their uses in indexing and searching are time-consuming. The space-filling curve that reduces the number of dimensions to one while preserving the neighborhood relation can be used in this context. A new fast technique for image indexing is developed which enables rapid insertions of new images without changing existing data. The retrieving is accelerated by avoiding the distance computing because images are ordered on 1-D data structure. Hilbert curve, the most neighborhood preserving space-filling curve, is used in the experimentation. A proposal of fast mapping facilitates the computing of 1-D Hilbert indexes from high dimensional features.

A new minimum trees-based approach for shape matching with improved time computing: application to graphical symbols recognition

Recently we have developed a model for shape description and matching. Based on minimum spanning trees construction and specifics stages like the mixture, it seems to have many desirable properties. Recognition invariance in front shift, rotated and noisy shape was checked through median scale tests related to GREC symbol reference database. Even if extracting the topology of a shape by mapping the shortest path connecting all the pixels seems to be powerful, the construction of graph induces an expensive algorithmic cost. In this article we discuss on the ways to reduce time computing. An alternative solution based on image compression concepts is provided and evaluated. The model no longer operates in the image space but in a compact space, namely the Discrete Cosine space. The use of block discrete cosine transform is discussed and justified. The experimental results led on the GREC2003 database show that the proposed method is characterized by a good discrimination power, a real robustness to noise with an acceptable time computing.

Alternative patterns of the multidimensional Hilbert curve: Application in image retrieval

Locality-preserving (distance preserving-mapping) is a useful property to manage multidimensional data. Close points in space remain -as much as possible- close after mapping on curve. That is why Hilbert space-filling curve is used in many domains and applications. Hilbert curve preserves well locality because from a construction aspect, it is guided by adajacency constraint on points ordering : the curve connects all points of a D-dimensional discrete space, without favoring any direction, under the constrainst that two successive points are separated by an unit distance. Originally defined in 2-D, all existing multidimensional extensions of the Hilbert curve satisfy adjacency by using the RBG pattern (based on Reflected Binary Gray code). The RBG pattern is then duplicated and arranged (geometrical transformations) to build the multidimensional Hilbert curve at a given order. In this paper, we emphasize that there are other patterns that can satisfy the adjacency. A formulation is given, an algorithm to find out solutions is provided and their respective level of locality preservation is estimated through a standard criterion. Results show that some new patterns can carry a comparable levels of locality and sometimes better than RBG. Moreover, selecting the best locality preserving pattern allows one to design, through orders, a new curve with a comparable overall locality preserving refer to Hilbert curve. The contribution of new patterns is experimented through a CBIR (Content-Based Image Retrieval) application. Large-scale image retrieval tests show that exploring the image feature space with an alternative way to the classical Hilbert curve can lead to improved image searching performances.

Proposition d’une famille de courbes remplissant l’espace de niveau de localité comparable à la courbe de Hilbert

RÉSUMÉ. Les courbes remplissant l’espace sont largement utilisées dans plusieurs domaines de l’informatique où la conservation de la localité est souvent considérée comme le critère pour choisir le type de courbe opérant dans une application. C’est pourquoi, la courbe de Hilbert, en qualité de courbe préservant au mieux la localité, reste majoritairement employée. Cependant, existe-t-il d’autres courbes vérifiant un niveau de localité comparable à la courbe de Hilbert ? Dans cet article, nous proposons une méthode flexible permettant de construire une famille de courbes remplissant un espace multidimensionnel de niveau de localité comparable à la courbe de Hilbert et parfois meilleur. Des tests expérimentaux comparatifs tendent à confirmer ces résultats. Notons que des éléments de preuve sur la validité de notre proposition sont avancés, sa mise en œuvre est illustrée à travers de nombreux exemples.

A NEW MINIMUM SPANNING TREE-BASED METHOD FOR SHAPE DESCRIPTION AND MATCHING WORKING IN DISCRETE COSINE SPACE

In this article, a new minimum spanning tree-based method for shape description and matching is proposed. Its properties are checked through the problem of graphical symbols recognition. Recognition invariance in front shift and multi-oriented noisy objects was studied in the context of small and low resolution binary images. The approach seems to have many desirable properties, even if the construction of graphs induces an expensive algorithmic cost. In order to reduce time computing, an alternative solution based on image compression concepts is provided. The recognition is realized in a compact space, namely the Discrete Cosine space. The use of block discrete cosine transform is discussed and justified. The experimental results led on the GREC2003 database show that the proposed method is characterized by a good discrimination power, a real robustness to noise, with an acceptable time computing. The position with a reference approach like Zernike moments is also investigated to measure the relevance of the proposed technique.