Gerhard WEBER, Thorsten PUCK, Univ.HARZ (Germany)
1.Detailed Abstract
raised-line drawings, tactile printing, gestures, moon, Braille
Printing Braille is a standard method for reproduction of Braille books, journals and press products. The quality of Braille depends on the paper and the scape and stiffness of the Braille dots.
Acceptance of Braille is among the general audience much higher than acceptance of electronic documents which may be read with paper-less Braille display or by listening to a speech synthesizer. But handling of electronic documents is increasing - maybe due to the widespread use of document reading machines also in the private domain.
When it comes to telecommunication both approaches meet. Paper-based documents are transferred using the scanner of a fax device and the printer of the receiver’s fax machine.
Using this technology for Braille readers is a new approach to utilize a otherwise very common and cost effective method of telecommunication.
We will describe in the following the implications of such utilization both in the domain of managing electronic documents and in the domain of handling graphics and braille on paper-bound media. Our implementation of TACFAX - a software application for handling faxes - tries to resemble these implications.
There are two different approaches to reproduction methods of tactile or relief material [1,5,6]:
Within project PRINT a print technology is developed by deposing specific ink in an ink-print process on paper. The resolution of this process is up to 300 dpi, while the levels of relief structures are variable. The speed of printing is comparable to that of a regular ink-jet printer. Earlier comparable products such as the Pixelmaster [2] where however less cost-effective and could not handle easily thick paper which makes reading braille more comfortable.
First test results of reading braille which were obtained by several blind subjects are encouraging and details are to be reported elsewhere.
This printing process not only allows printing of braille but also of Moon, a notation used for printing of books. The main advantage however is the composition of braille, moon and graphics within one sheet of paper. This gives the user the ability to relate text and graphs to each other.
4.FAX DOCUMENTS IN PRINT
Graphics appears in the context of textual material in fax documents. Current solutions to process fax documents are integrated with general purpose document reading machines. That is, the textual contents of a fax is read via speech synthesis or paper-less braille display. Even recordings of the synthesized voice are taken in order to handle the document through a more simpler cassette tape.
TACFAX takes these requirements into account. The key features of TACFAX are:
The nonvisual user interface of TACFAX itself is based on speech synthesis using, future versions may also include the usual features of a document reading machine.
5.ANALYSIS OF DOCUMENTS
We conducted a series of tests using business letters. The recognition of graphics by a standard OCR toolkit is less than optimal, although the recognition of text is nearly perfect . OCR is usually based on a histogram: a horizontal projection of black pixels for each scanning line of the page. Text lines result in an accumulation of pixels in the histogram. The number of pixels between text lines is ideally zero. Likewise a vertical histogram allows to identify tables. Graphic elements in documents are determined since they consist of a gray-scaled pixel matrix, which is reflected in both types of histograms.
These pixel accumulations allow to identify zones within a documents. The type (text, table, or graphics) of each zone however is to be determined by the absence of lines or columns, respectively .
Consequently we have investigated the recognition of graphics by different types of graphics like logo, emblem, graphics with colored background, hand written graphics, pictures or photographs.
Our tests show that logos or emblems of written text affect the process of graphic recognition because their histogram is similar to the histogram of text elements. Graphics with colored background including text, pictures and photographs are recognized as graphic element because they consist of a pixel matrix.
Depending on their motives, some special hand written graphics were recognized. A reason for misrecognition may be that a horizontal painted line is ‘confused’ with the underlining of text. Furthermore handwriting may not cause a striking value in histogram because of white background. Consequently it is rarely recognized.
Technical diagrams may include boxes, circles, arrows, etc and text for annotation. The textual zones in these samples should be inside the graphical zones, which was never detected by the OCR toolkit.
6.GRAPHICAL MARK-UP FOR DOCUMENT ANALYSIS
Based on our results we have developed a series of gestural mark-up commands for zones of a printed document [7, 8]. The automatically detected zones are tactile printed together with the document. Our approach allows the blind reader to identify zones by rectangular shape through their type, size and position. Figure 1 shows how three types of zones are distinguished.
Mark-up is accomplished through a regular mouse which is hold such that the users can still touch the tactile printout. Thereby a very cost-effective pointing device can be realized. A more sophisticated version may use a touch pad [3]. The complete version of the paper will report on first feedback from several users.
The actual movements are: rectangular shape to create a new zone, diagonal stroke from upper left corner to lower right corner of an existing zone to reassign their type and vice versa to delete a zone.
After zones have been identified properly the OCR process is improved and graphics is printed by TACFAX together with text.
7.ACKNOWLEDGMENTS
This work has been supported by the PRINT consortium, a research project of the EEC program "Telematics for the Disabled and Elderly" (TIDE).
8.REFERENCES
FULL PAPER
1.[ Abstract ]
2.[ Introduction ]
3.[ Tactile Reproduction ]
4.[ Fax documents in print ]
5.[ Analysis of documents ]
6.[ Graphical Mark-up for document analysis ]
7.[ Conclusions ]
8.[ Acknowledgments ]
9.[ References ]
Processing of fax documents is an option of current document reading machines which excludes the reader from all kind of graphics. Our tests show the erroneous detection of graphical zones by an industry standard OCR toolkit.
A new tactile printer which is developed within project PRINT generates high-resolution tactile output. We develop an interactive FAX system to allow the user to mark-up a tactile printed fax document using a pointing device for improving the OCR process, convert textual contents of a fax into Braille or Moon and combine graphics with tactile notations.
raised-line drawings, tactile printing, gestures, moon, Braille
Printing Braille is a standard method for reproduction of Braille books, journals and press products. The quality of Braille depends on the paper and shape and stiffness of the Braille dots. Acceptance of Braille is among the general audience much higher than acceptance of electronic documents which may be read with paper-less Braille display or by listening to a speech synthesizer. But handling of electronic documents is increasing - maybe due to the widespread use of document reading machines also in the private domain.
When it comes to telecommunication both approaches meet. Paper-based documents are transferred using the scanner of a fax device and the printer of the receiver’s fax machine. Using this technology for Braille readers is a new approach to utilize a otherwise very common and cost effective method of telecommunication.
We will describe in the following the implications of such utilization both in the domain of managing electronic documents and in the domain of handling tactile graphics [1,4] and Braille on paper-bound media. Our implementation of TACFAX - a software application for handling faxes - tries to resemble these implications.
It is, in the sense of the word, instrumental to tactile fax that a suitable, tactile output medium should exist to produce hard copies of faxes [3,9]. The most common way to represent tactile information is embossing on paper. It is difficult, however, to emboss a useful representation of graphics, due to the inherent low spatial resolution of embossing and the restricted capacity of paper to stretch into form. Swell paper is another possibility, but it requires producing an ink-print copy first, and its tactile quality is not sufficiently accepted among blind readers for reading books in Braille. Within project PRINT, a wide variety of technologies basically applicable to relief printing was investigated (such as [2]) to determine their suitability for creating tactile output such as Braille and tactile faxes.
A user needs analysis was performed in four European countries, involving well over 100 blind and sighted users, to determine which were the characteristics users expected a future tactile printer to have. At the outcome of both studies, it emerged that a device much resembling a modern office printer, using solid ink printing technology to create relief images, would best fit the users' needs while being technically feasible within reasonable means. It was also determined that there was a requirement for a machine that could be used for Braille, Moon (a tactile font used in Great Britain, which is based on Latin capitals and is easier to learn for the elderly blind or those with learning deficiencies), tactile graphics, and also visually readable output. It was important for users that the machine should be able to combine various methods of output on the same page. This set of characteristics makes it possible to apply fax reproduction to the new printer.
From the technological point of view, tactile printing requires the deposition of far greater amount of material on paper than visual printing. This influences the design of the tactile printer, which should be capable of quickly laying down a thick layer of ink to form a relief image. In fact many of the characteristics needed for tactile printing are contrary to what is needed for visual printing, making it difficult to simply adapt existing technologies.
A proof of concept study was based on a industry standard solid ink print head. However, the goal of printing one page per minute with this technology proved elusive and there would have been great legal and financial obstacles in the way of adapting the technology to better suit the needs for relief printing. Accordingly, an all-new print head was developed. It is based, like most inkjet print heads, on piezoceramic elements which create small droplets of liquid ink and eject them at a high rate - in this case, typically 4 kHz. Seventy such elements are stacked in a print head, making it possible to print Braille at fax resolution at a rate of about one page per minute.
Some important considerations in the design of the printer were accessibility, safety, and of course, output quality. In other words it should be no more difficult or more hazardous for a blind person to operate than for a sighted one. Accordingly, a number of sensors and safety catches are built in. It is ensured that ink is safe to handle, and does not smell or leave stains when handled. It also is free of any substances users might be allergic to. The inks are required to have certain qualities in both the solid and liquid states, such as the right surface roughness and stickiness, or lack of such, to make reading easy, a certain hardness and shear resistance, and some elasticity so they won't come off if the paper is bent while reading. They also need to have a high-enough melting point to withstand extreme summer temperatures, but not too high to be handled by the piezoceramic elements, and their viscosity and surface tension when liquid need to be within manageable limits. Another consideration was colouring. It is planned to offer coloured inks, to facilitate tactile reading for the visually impaired with a certain degree of residual vision, and also to enable the printer to combine tactile and visual output on the same page. An eighteen-month study identified a small number of candidate inks capable of meeting all the specifications.
With accessibility in mind, the design team made sure there are clear and concise voice messages available for every printer state, hazardous condition, or error, that may occur during operation, such as low ink reservoir, open cover, paper out or jammed, etc. These messages will be available in most EU languages so users don't need to rely on a foreign language when operating the printer. The built-in control program allows operation and configuration either remotely from a PC or locally, using a built-in key array. Again, all possible settings and modes can be read out by the voice output.
As for fax integration, there will be a provision allowing the integration of the fax modem with the printer as a standalone fax machine, though the typical mode of operation would be to use it in connection with a PC.
4.FAX DOCUMENTS IN PRINT
Current solutions to process fax documents are integrated with general purpose document reading machines. That is, the textual contents of a fax is read via speech synthesis or paper-less Braille display. Even recordings of the synthesized voice are taken in order to handle the document through a more simpler cassette tape. For everyday use there is little assistance by the document reader to organize or classify paper documents. Moreover, graphics as they appear in the context of textual material in fax documents are ignored by document readers.
TACFAX takes these requirements into account. The key features of TACFAX are:
unattended processing of faxes,
translation of fax documents into Braille or moon using an industry standard OCR toolkit [6],
document management on the basis of HTML, also for users applying TACFAX in the private domain,
mark-up of graphical contents by the blind user, and
processing of graphics according to mark-up.
For an unattended reception of faxes TACFAX uses Microsoft Exchange. Incoming faxes are stored, processed and printed out immediately in Braille or moon [5]. A state of the art OCR toolkit determines sections (zones) of text, tables and graphical elements in a fax document. Recognised text is printed characater by character either
in a TrueType font for Moon notation, or
individual circles are drawn according to a Braille code definition file.
Any 6- or 8-dot Braille code can be applied (see Figure 1) and Braille is drawn with standard or non-standard dot-to-dot distance, dot diameter, or character to character distance.
.
Figure 1: Braille and Moon printout
Recognised graphics is printed at the end of the document. An image processing module includes routines to modify recognised graphics upon user request. Modifications allow to zoom, rotate, change contrast or remove noise. As the recognition of graphics is not completely error-free the selection of zones can be improved by the user, we will refer to this in the following as the mark.up process. (see following sections 5 and 6). For those users who want to work on the fax documents on a PC, HTML was chosen as document encoding. The HTML file includes links to the original image file (of formats JPG and TIF) and a link to a text file including the recognised text.
4.1The Nonvisual User Interface of TACFAX
The user interface of TACFAX is designed for three stereotypes of users:
novice,
advanced, and
expert.
Thereby TACFAX offers to bridge the gap between the mental model of a stand-alone fax receiver and electronic document processing. The task of a telephone operator may be to determine the receiver of the fax within the company, while a sales person has to base a quote on the actual figures listed within a fax and hence does not want to retype the numerical text.
The novice user uses the fax system as stand alone fax device - received faxes will be printed out immediately, graphics will be printed automatically at the end of the document. Faxes are printed in one of the selected notations (Braille, Moon or ink print, see Figure 2).
Figure 2: Flow diagram of TACFAX with default settings
In order to replicate the acoustic presentation of a fax machine we have implemented a series of nonverbal sounds which announce reception of a fax by a ringing phone, or indicate the processing of the fax by a machine sound (samples are found at http://www-informatik.fh-harz.de/print/PrintProjectEng.html). The only input which a user may have to generate is targeted at the printer itself, such as acknowledging reloading of paper or adjusting the thickness of a printout.
The advanced user not only receives faxes but may forward faxes. The user has to accept the concept that the fax is no more only a paper-based document but an electronic document. According to his reading results he can make use of the mark-up facilities to modify the electronic document. Therefore a tactile printout of an image of the complete fax document is printed. In addition to mark-up facilities the basic settings such as language and type of layout (letter, newspaper) for improved OCR recognition are changeable (Figure 3).
Furthermore the advanced user identifies faxes by type and assigns two keywords to each fax (a general subject and a subject). These keywords will be printed at the top of each page. With this method faxes from the same phone number will be assigned the same keyword (general subject). After mark-up, the print job will be started again.
Figure 3: Flow diagram of TACFAX with advanced settings
Attempts to automatically classify a fax document are beyond the scope of this work, although they may lead to sufficient results.
The nonvisual user interface for an advanced user is based on input through a stand qwerty-keyboard and speech output. While speech playback is suitable to announce digits and command options we have selected speech synthesis for verbal acoustic feedback in order to handle arbitrary categories of subjects and general subjects. Further field tests will show if this assumption can be simplified and simple speech playback is sufficient.
In addition to this, the expert user can retrieve and reprint older faxes by name due to file-based handling of faxes (see Figure 4).
Figure 4: Flow diagram of TACFAX with expert settings
The metaphor of the user interface is based on a word processor. TACFAX therefor includes features for connection with a word processor through HTML files and the direct networking features of Windows 95. Files can be ‘seen’ and be transferred to a connected desktop PC.
In summary, sources for processing of fax documents are the telephone system, a electronic copy of a fax image or previously received fax images. The processing of faxes is driven by various settings which allow to modify the recognition of documents and which allows to generate different tactile presentations. Our approach to improve document processing is thus both automatic for nocie users and semi-automatic for advanced and expert users who are blind. We are not specifically aiming at sighted users in order to explain graphics although the use of HTML allows to annotate some graphics by text or spoken input. A hypertext link within the fax document may lead to such annotation, but requires a HTML browser for the blind reader and an HTML editor for the sighted collegue, teacher, etc who authors some comment on the graphics.
5.ANALYSIS OF FAX DOCUMENTS
There are several different types of fax documents depending on their language and their layout. The types of layout are business letter (single columned), or newspaper article (multi-columned).
In addition to this some fax documents include graphics as well as a mixture of text and graphics. The documents which we are addressing allow to separate graphics from text by identifying disjunct areas (zones) in the fax image. For the analysis of documents an unobjectionable handling of recognition is inevitable. The recognition by a standard OCR toolkit must be based on the above options but handles the following document features automatically:
detection of page orientation position,
recognition of zones,
text recognition, and
spelling correction
After defining the layout and the language of the document by the user, the document is analysed by OCR. The OCR process starts by detecting and correcting the page orientation position. The automatic creation of zones separates text, tables and graphics and determines the individual zones with their respective type as well as the sequence of reading. The sequence of reading authoritatively depends on a predefined layout of document as shown in Figure 5.
original page layout sequence single-columned sequence multi-columned
Figure 5: Sequence of reading
Because each language includes special fonts, the text recognition bases on the predefined language. After recognition erroneous words are indicated and corrected by the use of a language-specific dictionary. At the end of the process, the recognised document is build.
We conducted a series of tests using business letters. The recognition of graphics by a standard OCR toolkit is less than optimal, although the recognition of text is nearly perfect . OCR is usually based on a histogram: a horizontal projection of black pixels for each scanning line of the page. Text lines result in an accumulation of pixels in the histogram. The number of pixels between text lines is ideally zero. Likewise a vertical histogram allows to identify tables. Graphic elements in documents are determined since they consist of a grey-scaled pixel matrix, which is reflected in both types of histograms.
These pixel accumulations allow to identify zones within a documents. The type (text, table, or graphics) of each zone however is to be determined by the absence of lines or columns, respectively .
Consequently we have investigated the recognition of graphics for different types of graphical input such as logo, emblem, graphics with coloured background, hand written graphics, pictures or photographs.
Our tests show that logos or emblems of written text affect the process of graphic recognition because their histogram is similar to the histogram of text elements. Graphics with coloured background including text, pictures and photographs are recognized as graphic element because they consist of a pixel matrix.
Depending on their motives, some special hand written graphics were recognized. A reason for misrecognition may be that a horizontal painted line is ‘confused’ with the underlining of text. Furthermore handwriting may not cause a striking value in histogram because of white background. Consequently it is rarely recognized.
Technical diagrams (compare with figures 2, 3 and 4) may include boxes, circles, arrows, etc. and text for annotation. The textual zones in these samples should be inside the graphical zones, which was never detected by the OCR toolkit. The tactile reading of such simple graphical shapes is trained during mathematical education within geometry classes. Hence we focus in the following on a method which allows to improve recognition of fax documents which include technical diagrams.
6.GRAPHICAL MARK-UP FOR DOCUMENT ANALYSIS
Automatically detected zones are printed tactile within the image of document. Our approach allows the blind reader to identify zones by a rectangular shape through their type, size and position in the context of an image of ink-print text. The cognitive model of the document layout can be constructed by touching the tactile version of the fax. As such, the spatial properties of text or the context of text can be determined. The user is not asked to actually read text in ink-print, but to isolate text and graphics from its tactile perception. Figure 6 shows three types of zones: text, graphics and tables. They are distinguished by simple geometrical shapes, namely box, box with diagonal lines and grid.
Figure 6: Zonal mark-up: text, graphics,and tables
A blind user may judge the type of a zone better than the OCR, even if it is presented in a tactile manner. Identification of zones and their type is possible through a simple command based interface where each zone is assigned a number. However repositioning a zone and re-sizing it is very unnatural if the user has to rely on numerical expressions. A deictic input method greatly improves the use of this facility.
Based on our test results we have developed a series of gestural mark-up commands for zones of a printed document [7, 8]. Mark-up is accomplished in our first experiments through a regular mouse which is hold such that the users can still touch the tactile printout. Thereby a very cost-effective pointing device can be realized. A more sophisticated version may use a touch pad [3].
6.1 GESTURAL INPUT
Gestures are an interaction method which require no immediate visual feedback [7]. They are therefor suitable as an input method for blind users [10]. The active perception of an upper limb’s movement is sufficient to form a unique shape, that is a path of the hand or finger which can repeated and which conveys some semantic expression . However gestures - unlike everyday sounds - have to be learned as their use is different from person to person[12]. They tend also to be forgotten after several weeks of not using them or an alternative path is used to convey the same semantic expression. By applying gestural input to a combination tactile presentation and touch senstive input a combination of form-based deictic input and tactile guidance of the gestural movement can be achieved [11].
Current approaches to deictic input try to minimize the number of gestures and apply them
to pointing tasks and direct manipulation interaction methods e.g. drag and drop or
to basic text operations such as delete (cross) and select (circle).
Gestures on the basis of a pointing device are also preferably formed with one stroke in order to more easily distinguish start and end of gestural input, although gestural input based on a data glove may be separated through detection of the speed of movement of individual fingers.
Pointing by blind people is in general accepted by blind people according to an evaluation of Fitt’s Law in the context of adapting graphical userfaces [Petrie] but effectiveness depends on many technical considerations.
Mark-up of zones is a combination of a pointing task with straight linear movement and sequences of linear movement (see Table 1 for some samples of gesture descriptions and operations).
| gestural input | Operation |
| a horizontal line within a boxed zone | assign the type „text" to a zone |
| a diagonal (upper left to lower right) line | assign the type „graphics" to a zone |
| Four lines which are either horizontal or vertical and whose start is „close" to the end (box) | create a zone |
| a cross of two diagonal lines | delete a zone |
Table 1: Gestural input and operations
These movements on the document specify at the same time the operation to be performed and the position of a zone. Resizing a zone and combining two zone is a sequence of operations delete and create a new zone.
6.2 FEEDBACK FOR GESTURAL INPUT
An informal evaluation of this input method was undertaken with tactile printouts made by the proof-of-concept printer prototype and actual fax documents that were processed by the TACFAX system. Users reported that tactile information about zones was to much covered by the tactile image itself. Figure 7 describes a sample document with erroneous detection of graphical zones as used during evaluation. While Tour d’Eiffel was clearily recognized as graphics in Figure 7, part of an logo is marked as a separate zone but of type text. The blind reader should re-assign the type to graphics.
Figure 7: Sample mark-up of a document
A re-design of gestural input is now including verbal acoustic feedback which identifies the type of a zone as soon as the user reaches it. Thereby the effect of „getting lost" after crossing the box boundaries (which are marked tactile) is counter-acted, although speech stops after a while and continues only to announce leaving the zone.
In the context of expressing mathematical equations the possibility to provide an glance of a structure has proved to be effective [8]. We therefor plan to describe the zones verbally upon user request. This seems suitable as long as are no overlapped zones. However, graphical mark-up allows specifically to place a textual zone within a graphical zone. Verbal description in this example needs to be further improved.
7. CONCLUSIONS
Document processing for non visual presentation can be based on a combination of both auditory and high-resolution tactile presentation. Through interactive analsysis of documents the recognition of text and graphics can be improved. Instead of an automatic method a semi-automatic method can increase accessibility to fax documents.
Having started in January 1997, PRINT will run through December 1999, by which time it is intended to have completed a one-year field study with a certain number of printer demonstrators, part of which will feature the tactile fax system with graphical mark-up.
We want to thank Prof. DR. Giessner and Reinhold Haiß, Fachhochschule Augsburg for preparing the tactile samples. This work has been supported by the PRINT consortium, a research project of the EEC program "Telematics for the Disabled and Elderly" (TIDE).
9. REFERENCES
Gerhard WEBER, Thorsten PUCK, Univ.HARZ (Germany)
Max LANGE, BLISTA BRAILLETEC GMBH, (Germany)
1.1. Mots-clés
dessins en relief, impression tactile, gestes, gros caractères, braille.
INTRODUCTION
L'impression en braille est une méthode classique de reproduction des livres braille, des journaux et des produits de la presse. La qualité du braille dépend du papier ainsi que du relief et de la rigidité des points braille.
Le braille est beaucoup mieux accepté du grand public que les documents électroniques pouvant être lus en braille éphémère ou en écoutant une synthèse vocale. Mais le maniement des documents électroniques est en plein essor, peut-être parce que l'usage des machines à lire les documents se répand également dans le domaine privé.
Dans le domaine des télécommunications, les deux approches se rejoignent. Les documents papier sont transférés au moyen du scanner d'un fax et de l'imprimante du fax du destinataire.
Utiliser cette technologie pour les lecteurs de braille est une approche nouvelle de l'emploi d'un procédé de télécommunication qui est par ailleurs très courant et peu coûteux.
Nous décrirons ci-dessous les implications d'une telle utilisation aussi bien dans le domaine de la gestion des documents électroniques que dans celui de la manipulation des graphiques et du braille sur le papier. Notre mise en oeuvre de TACFAX, un logiciel de gestion des fax, s'efforce de réagir à ces implications.
LA REPRODUCTION TACTILE
Il existe deux approches différentes des procédés de reproduction des matériaux tactiles ou en relief (1,5,6) :
1) en déformant du papier, une feuille de plastique, ou même du bois ;
2) en ajoutant des matériaux à un substrat tel que le papier.
Dans le cadre du projet PRINT, une technique d'impression a été mise au point en déposant une encre spéciale sur du papier dans un processus d'impression à l'encre. La résolution peut ainsi atteindre 300 dpi, tandis que les niveaux des structures en relief sont variables. La vitesse d'impression équivaut à celle d'une imprimante à jet d'encre ordinaire. Les produits comparables existant précédemment étaient toutefois plus coûteux et ne pouvaient pas utiliser facilement du papier épais, qui offre un meilleur confort de lecture du braille.
Les premiers résultats des tests de lecture du braille obtenus par plusieurs sujets non-voyants sont encourageants, et les détails en seront rapportés ailleurs.
Ce procédé d'impression permet non seulement l'impression du braille mais aussi des gros caractères, un système de notation utilisé dans l'impression des livres. Le principal avantage demeure cependant la composition du braille, des gros caractères et des graphiques sur la même feuille de papier. Cela permet à l'utilisateur d'établir le lien entre le texte et les graphiques.
LES DOCUMENTS FAX SOUS FORME IMPRIMÉE
Les graphiques apparaissent dans les documents fax dans un contexte textuel. Les solutions actuelles de traitement des documents fax sont intégrées dans les machines à lire les documents à usage général, c'est-à-dire que le contenu textuel d'un fax est lu au moyen d'une synthèse vocale ou d'un affichage braille éphémère. On enregistre même la synthèse vocale afin d'accéder au document sous la forme plus simple d'un enregistrement sur cassette.
TACFAX tient compte de ces exigences. Les principales caractéristiques de TACFAX sont :
- le traitement autonome des fax,
- la conversion des documents fax en braille ou en gros caractères au moyen d'un outil de reconnaissance optique de caractères classique du commerce (4),
- la gestion de documents sur la base du HTML, également pour les utilisateurs employant TACFAX dans le domaine privé,
- le balisage du contenu graphique par l'utilisateur non-voyant, et
- le traitement des graphiques en fonction du balisage.
L'interface utilisateur non visuelle de TACFAX lui-même est basée sur l'utilisation de la synthèse vocale, et il pourra s'y ajouter, dans les versions ultérieures, les caractéristiques habituelles d'une machine à lire les documents.
L'ANALYSE DES DOCUMENTS
Nous avons procédé à une série de tests avec des lettres à caractère commercial. La reconnaissance des graphiques par un outil de reconnaissance optique classique laisse beaucoup à désirer, quoique la reconnaissance du texte soit presque parfaite. La reconnaissance optique se base généralement sur un histogramme : une projection horizontale de pixels noirs pour chaque ligne de la page à scanner. Les lignes de texte sont le résultat d'une accumulation de pixels dans l'histogramme. Le nombre de pixels entre les lignes de texte devrait idéalement être nul. De même, un histogramme vertical permet l'identification des tables. Les éléments graphiques des documents sont identifiés parce qu'ils prennent la forme d'une matrice de pixels à niveaux de gris, qui est reflétée dans les deux types d'histogrammes.
Ces accumulations de pixels permettent l'identification des zones dans un document. Le type de chaque zone (texte, tableau ou graphique) doit toutefois être déterminé par l'absence de lignes ou de colonnes respectivement.
En conséquence, nous avons étudié la reconnaissance des graphiques avec différents types de graphiques tels que les logos, les symboles, les graphiques à fond coloré, les graphiques écrits à la main, les dessins ou les photographies.
Nos tests montrent que les logos et les symboles présents dans le texte écrit affectent le processus de reconnaissance parce que leur histogramme est semblable à celui des éléments de texte. Les graphiques à fond coloré comprenant du texte, des dessins et des photographies sont reconnus comme des éléments graphiques parce qu'ils prennent la forme d'une matrice de pixels.
Selon leurs motifs, certains graphiques particuliers écrits à la main ont été reconnus. L'une des raisons de leur mauvaise reconnaissance est peut-être qu'une ligne horizontale dessinée est "confondue" avec du texte souligné. En outre, du texte écrit à la main peut ne pas produire de valeur nette dans un histogramme en raison du fond blanc. C'est pourquoi il est rarement reconnu.
Les diagrammes techniques peuvent comprendre des rectangles, des cercles, des flèches etc. et du texte d'annotation. Les zones texte de ces échantillons se trouveraient à l'intérieur des zones graphiques, ce qui n'a jamais été détecté par l'outil de reconnaissance optique.
LE BALISAGE DES GRAPHIQUES POUR L'ANALYSE DES DOCUMENTS
Sur la base de nos résultats, nous avons mis au point une série de commandes de balisage gestuel pour les zones d'un document imprimé (7,8). Les zones détectées automatiquement sont imprimées sous forme tactile en même temps que le document. Notre approche permet à l'utilisateur non-voyant d'identifier les zones de forme rectangulaire par leur type, leur dimension et leur position. La figure 1 montre comment ces trois types de zones sont distinguées.
Figure 1 : Balisage des zones : texte, graphiques, tableaux
Le balisage s'effectue avec une souris ordinaire tenue de façon à ce que les utilisateurs puissent toujours toucher le document imprimé sous forme tactile. On peut ainsi réaliser un dispositif de pointage très peu coûteux. Une version plus élaborée pourrait utiliser un pavé tactile (3). La version complète de cet article rendra compte des premières réactions de plusieurs utilisateurs.
Les mouvements sont en réalité : une forme rectangulaire pour créer une nouvelle zone, un trait en diagonale du coin en haut à gauche au coin en bas à droite d'une zone existante pour en redéfinir le type, et l'inverse pour effacer une zone.
Une fois les zones correctement identifiées, le processus de reconnaissance optique est amélioré et les graphiques sont imprimés par TACFAX en même temps que le texte.
REMERCIEMENTS
Ce travail a bénéficié du soutien du consortium PRINT, un projet de recherche du programme européen "La télématique pour les handicapés et les personnes âgées" (TIDE).
RÉFÉRENCES
[Actes]
