Recognition of Amharic Braille
No Thumbnail Available
Date
2009-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
According to the Il1Iernational E)'e Foundation (IEF) reports, there are currently about45 million visually
impaired people in the world, the vast majority of which has been living in Africa In Ethiopia, the latest
census indicates that there are well over half a million visually impaired individuals including: studel1ls.
lawyers. teachers, researchers, artists etc. So far, Braille has been the most invaluable means for visually
impaired individuals to communicate to the world. Braille called after its inventor Braille Luis is a tactile
writing means that consists of six dots arrangement in 2-by-3 matrix in a cell. Societal support is
important to rebuild lives devastated by sight loss. These days technology has contributed a paramount
value 10 the society in facilitating two way communications. Recently, optical character recognition
(OCR) technique has been implemented for the recognition of Braille documents scanned with standard
scanning device.
In this study, an attempt has been made in Amharic Braille-Io-print documents recognition. To achieve
this, various techniques has been reviewed, developed and adopted. The proposed system performs the
required recognition in two phases: these are recognition of Braille character and Braille-to-print
character. The first phase involves four steps such as threes holding/Linearization, image-segmentation, and
feature extraction and recognition. Global-threshold has been implemel1led to binaries the foreground
(col1len/) from the background. As Braille cell are strictly arranged horizontally and vertically, mesh-grid
technique has been adopted for segmentation process. With mesh, dots are extracted following the
vertical and horizontal grid line. Having done this, in feature extraction the system once again lakes
advantage of the mesh. So, during this sleeps dots are farther grouped in to cell, which would then
recognized with context analysis based on rules defined. To accomplish the first phase Microsoft Visual
C++ programming tools has been used.
The second phase, deals with classification of Braille-to-print. To this end MA TLB 's implemel1lation 0/
The feed forward artificial neural network has been utilized.
The neural network classifier has been trained on Amharic Braille with Amharic print as the target
character. lV1oreover, the performance of the model has been evaluated with test sets that are prepared
.from the Braille document
Eventually,, the study has shown better performance with all training and lest set, with 92.5% accurate.
Description
Keywords
Information Science