attached
Recent technology advancements in high performance processing hardware and
intelligent learning algorithms have caused an up-raise in technologies that aims to
mimic human behavior. When it comes to a visually impaired person they can't read
a context of a book or a paper as an ordinary human being. In this research work, an
attempt is made to develop off-line or online recognition strategies for the isolated
printed English character (A to Z) recognition in order to help visually impaired
people to overcome their disability in reading. Machine learning algorithms can be
trained to recognize specific letter shapes and sizes because in printed materials the
font, size and the features of letters are comparably similar to each other. For word
recognition systems, variability in the form of noise caused by imperfection in letters,
image acquisition methods and background noise has become the most challenging
issue for performance. However the uses of neural networks have shown successful
results in yielding comparably high recognition accuracy levels.
The proposed system uses a real time images processed by MATLAB to identify the
characters. Images are obtained in specific formats such as JPEG, BMT etc. Then the
noise which comes to the image in acquisition is filtered. After that colored image is
converted to gray scale and the identified image is then convefted to a binary image
using a suitable threshold. Then blob detection is used to identify connected areas
and bounding boxes are used to identify individual letters. Finally recognition is done
using a Statistical algorithm which uses a pre stored example database of English
alphabet and saved pictures. Apart from the recognition paft a complete system of
image acquisition and a text to voice conversion strategy is developed to make the
system more useful. Moreover the developed system which unlike prevailing systems
has the ability to not only recognizes characters but also to speak the recognized
words for the user. Another advantage of the proposed system is that it can be
implemented using off-the-shelf components like raspberry pi and associated image
acquisition devices which results in minimizing overall cost of the system.
