Size Independent Bangla Character Recognition System. Using Artificial Neural Network

      Аннотация: The character recognition process is implemented in the two stages. They are learning stage and testing stage. In learning stage, the following steps are considered. These are Scanned Character by scanner, Pre-processing, Feature Extraction, Training and a set of processed character. At first, the character to be recognized is scanned through a scanner and thus the character becomes a bitmap image. Then the boundary region of the saved image I is extracted. The feature of the scaled image is extracted and converted into an m x n matrix, which is reduced to very small matrix such as 16 X 16 using by matlab imresize scaling function. The feature matrix is then fed to the multilayer feedforward Neural Network. The testing stage includes Scanned Character, Pre-processing, Feature Extraction, Classification and Classified Character. In the phase of Classification, the selected features have been contained enough information within it to identity each character class uniquely. At the last phase, each unknown characters are tested through the experimental extracted features of character. The Bangla ‘Banjonborno’ and ‘Sorborno’ characters have been chosen to test the recognition System.

Ключевые слова: Feature Extraction, classifier, Adaptation method, Generalized delta rule, Multilayer feedforward neural network