A Comparison between Support Vector Machine (SVM) and Bootstrap Aggregating Technique for Recognizing Bangla Handwritten Characters

This paper represents the optical character recognition for Bangla handwritten characters using the popular classifier SVM and Bootstrap Aggregating technique. The segmentation process in Bangla is difficult because of complex letters and "Matra (top horizontal line)" in the words. For the feature extraction method there was no particular algorithm found, which was efficient enough, so in this experiment the Hog feature extraction and Binary pixel feature extraction methods were used. Hog features and Binary pixel features were combined for the proposed system. To recognize a character Support Vector Machine (SVM) and Bootstrap Aggregating were used. Experimental results for the SVM classifier and Bootstrap aggregating shows 100% accuracy for trained characters and for random untrained characters, SVM classifier shows accuracy about 89.8% and for the Bootstrap Aggregating method the accuracy is 93%.