Abstract:
This research applied a genetic algorithm in the pattern of cell automata
and through Conway s rules of the game of life, to generate a system of printed Thai
character recognition. The system consisted of two main parts, namely, recognition
training and recognition testing. The printed character images fed to the first part were
derived from standard character patterns widely used in a computer currently totalling
72,864 characters. As for the images used for recognition testing, they were captured
from a computer screen and stored in BMP pattern, amounting to 1,015 characters.
The findings in this research revealed that the database used was of
large size and data was transformed from a table frame of 64 x 64 pixels to be stored
in the form of bit strings. A table size of 64 x 64 pixels was used to enable a wide
variety of distribution patterns of the stable state of each character, making its identity
more obvious. This, of course, caused a modification process in each generation till
the final generation which took a long time while the database was used to represent
the population of the final generation of each character must be large enough for the
bit string used to represent these characters. This would enable the system to recognize
a character based on its frequency with the largest number of those bit string patterns.
Out of 1,015 printed Thai characters tested, it was found that the system
could recognize (accept) 986 characters or 97.14%, while rejecting 6 characters or
0.59% and misrecognizing 23 characters or 2.27%. The recognition speed is 85
seconds per character on the average.