tools can also be used to print the Braille output in black on a laser printer for proofreading by sighted users.

•The drawback is due to slight incompatibilities between the computer code (ASCII) and the Braille code which leads to additional ambiguities. For example, the single ending quotation mark and the apostrophe are the same in ASCII, but different in Braille (Gray 1998).

Figure 14.20 does not include conversion of the contents of a computer screen or web documents. Reading the screen contents is the main application of Braille displays. As this chapter is concerned with speech and language, the related issues of interpreting layouts, embedded graphics and formulae will not be discussed and the reader is referred to Chapter 13. A discussion of the different formats which can be used to represent the input of the conversion process (ASCII, MS Word, LaTeX, HTML, and SGML) have been given by Kahlisch (1998).

There are differences in the software required according to whether the Braille output is directed to an embosser or to a Braille display. These details are not considered in Figure 14.20. When combined with a computer keyboard, a Braille display should be able to switch between the following modes:

•Navigation (exploring the contents of the screen).

•Communication with a program.

•Reading long texts.

14.5.3 Braille-to-text Conversion

The (re)conversion of Braille to text is required in a number of applications. The main application is communication by public organizations with blind people. Other applications are in education and research, where sighted people may want to read Braille documents. Automated Braille-to-text conversion does not require any special hardware, as a standard flat-bed scanner can be used to input the Braille symbols and further processing carried out on a PC. The standard OCR software for the scanner needs to be replaced by software that is able to recognize Braille symbols. After recognition by the OCR software, the sequence of Braille symbols is transformed to the corresponding sequence of text characters. The Braille recognition and transformation software will be described in more detail below.

Another application of Braille-to-text conversion, which only requires the second (transformation) step, is the conversion of Braille files into text files for subsequent text processing or other electronic applications. Examples include production of the minutes of a meeting using a portable electronic Braille notetaker. The Braille minutes are transferred from notetaker to a PC which produces a Braille text file which must be converted to a text file before it can be distributed to sighted partners by e-mail or in paper format.

536 14 Speech, Text and Braille Conversion Technology

OCR software

Other than in text reading, Braille symbols embossed on a thick sheet of paper cannot be distinguished from the background by their colour and the Braille dots must be identified by their appearance in the light of the scanner lamp. Many Braille documents are in “interpoint Braille”, with Braille symbols embossed from both sides of the paper to utilize the voluminous carrier material more effectively. This requires the OCR software to distinguish between the raised bumps and depressions. This poses additional challenges compared to the text-based OCR presented in Chapter 15. The essential steps in the recognition process include the following (Antonacopoulos 1997):

•Image acquisition. The camera obtains a grey-level image with each dot represented by a light area combined with a shadow. The details of the image will depend on the type of scanner and its angle of illumination. A spatial resolution of between 80 and 200 dots per inch is recommended.

•Pre-processing. The image quality is improved using image processing methods.

•Identification of the dot. The software identifies the regions of the grey-scale image which correspond to a dot.

•Segmentation of the Braille symbols. The detected dots are grouped as Braille symbols or cells. The decision on which points belong to the same cell depends on the vertical and horizontal spacing between dots. This is not an easy task as can be seen, for instance, by considering the horizontal spacing. The following distances a, b, and c are of increasing size and defined as the distance between the two columns of one symbol, a, distance between two adjacent symbols, b, and the distance between two adjacent words, c. If a symbol with dots only in

its left column is followed by a symbol with dots only in its right column, the distance is (2 a + b) and this distance most be interpreted appropriately.

•Recognition. The positions of the symbols (cells) are known from the previous stage. The cells must be assigned a particular Braille symbol according to the presence or absence of dots in the six possible positions of the cell.

•Verification. Recognition errors can result from a number of different factors. Braille code does not have any redundancy at symbol level, as all possible combinations of dots are allowed. Therefore consideration of the context is required to verify the results of Braille recognition. This task can be carried out by the conversion module.

It should be noted that Braille-to-text conversion devices achieve surprisingly high recognition performance for good quality documents, whereas problems may occur for older or dirty Braille documents.

Conversion software

To conclude this section, two tasks of the module following the OCR software are discussed briefly: