Recognition of Double Sided Amharic Braille Documents
No Thumbnail Available
Date
2015-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Amharic language has large number of characters. As a result, Amharic Braille image recognition into print text is not an easy task. Amharic Braille cell formulation, encoding to a Braille code and translating the code to print text are different from Braille recognition systems of foreign languages’ characters. Few researches have been conducted in recognition of Amharic Braille documents. However, recognition of double sided Amharic Braille documents, which needs segmentation and identification of recto and verso dots from the background, and separation of overlapping recto and verso dots, has not been conducted so far.
In this work, we propose a design for recognition of double sided Amharic Braille documents. The design has a preprocessing, segmentation, dot identification, page formulation, transformation, and recognition components. We used direction field tensor to preprocess and segment dots from the background. After segmentation, gradient field is used to identify a dot as recto or verso. Overlapping dots were further segmented and identified using Braille dot attributes (centroid, orientation, and area). The identified recto and verso dots are separated into two separate images or pages using the page formulation component. Then, we used Braille cell encoding algorithm in order to formulate identified recto or verso dots into a Braille code. Finally, the Braille code is translated to print text using Braille code translation algorithm. The designed algorithms encode and translate the dots starting from left-top corner of the first dot to the right downward over the page. In order to use the same Braille cell encoding and Braille code translation algorithms for both pages, dots on the recto page are mirrored about a vertical symmetric line. Moreover, we used rotation in reversing wrongly scanned documents automatically as long as the translation performance is less than some threshold value which notifies the system the page is wrongly scanned.
In order to test the proposed design, we developed a prototype using MATLAB and test performances of dot identification and translation of double sided Amharic Braille images to print texts. We achieved an average dot identification accuracy of 99.3% and average translation accuracy of 95.6%. This is remarkably motivating performance as it is the first achievement in recognition of double sided Amharic Braille documents.
Key Words: - Braille Cell, Direction Field Tensor, Gradient Field, Recto Dot and Verso Dot.
Description
Keywords
Braille Cell; Direction Field Tensor; Gradient Field; Recto Dot and Verso Dot.