Abstract:
Recognizing handwritten characters is a major challenge in the field of pattern recognition. A unique solution is difficult to find because of the high variability in the samples.
Although a lot of research is going on to identify Bangla handwritten characters, the developed methods cannot be universally applied due to the lack of a central database. Each researcher has to manage their own data, which creates non-uniform results.
Artificial Neural Network (ANN) is a well-established method in the field of pattern recognition for recognizing handwritten character. Various ANNs have been developed to identify handwritten characters in different languages. Here, we use ANN with back propagation learning algorithm to classify Bangla vowels.
The data used to identify the characters was collected by us and was fed to the neural network after preprocessing.Using the pixel values as data, the best result obtained was 68.9% recognition rate for 16 hidden layers, which is quite poor. To improve the results, we used a Gabor filter to extract directional features from the character images.
With the feature data, the best result obtained was for 207 hidden layers, which is 79.4%. Finally, the drawbacks and future works are briefly discussed.
Description:
This thesis submitted in partial fulfillment of the requirements for the degree of B.Sc in Electrical and Electronic Engineering of East West University, Dhaka, Bangladesh.