| Summary: | When computer technology progression was reached in 20th century, many new branches of computer science developed based on this computer technology development. Artificial intelligent is one of many new branches that developed.
Artificial neural network is one of many branches of artificial intelligent that investigate human brain neural in computer, it is hoped that computer can do human tasks, such as classification, assoiation, optimization and self organizing. Neural network built is only the replica of human brain neural network, which is limited and impossible to compete with human brain. But then, the problem lays on how the system can be built so it can do a good job.
This research is to develop the ANN system to recognize human's handwritten characters with various deformation. The system makes use of the 25 x 20 pixel matrix cell to represent a single character. This matrix cell is grouped into 5 x 5 pixel submatrix, resulting another 5 x 4 new matrix. The element of the last matrix is the number which representing the number of pixels covered by the character being recognized. The 5 x 4 matrix then is multiplied with (0 1 1 0) vector, which resulting a vector as an input of the ANN.
The ANN recognizing algorithm used here is the back-propagation combined with supervised learning method. The rate of learning is given 0.05 with error tolerance 0.01. The experimental results show that the system performs quite well with 86,9% average accuracy.
Keywords : artificial neural network, back-propagation, matrix cell, supervised learning, target error
|
|---|