Abstract:
The Braille system is the most valuable and indispensable method that enables
blind or visually impaired people to write and read through the concept of the
Braille cell. Braille is a great help in bridging the written communication gap
between the blind and braille-illiterate people. Also, written materials used by
the visually impaired can only be read by those proficient in Braille, so there
is no precise way for ordinary people to assist the visually impaired in written
communication. This study aimed to develop a model for ordinary people who
do not know the Sinhala Braille system to understand the Sinhala Braille. The
image datasets of 55 Sinhala Braille characters were collected using a mobile
phone camera to develop this model. All the input images were resized to
28X28 pixels during the preprocessing process. After, the images are binarized
with local adaptive thresholding. Then, those images were subjected to
procedures such as grayscale, histogram normalization, gaussian filter,
threshold binarization, erosion, and dilation. Finally, the preprocessed images
were fed into a well-trained Convolutional Neural Network model. The
developed model tested for 55 Sinhala Braille characters with eight
punctuation marks and achieved an overall accuracy of more than 97%. The
model will be further developed as a simple mobile and web application to
overcome the limitations of written communication between the blind and
Braille illiterate sighted people in Sri Lankan society.