Quad-Tree Based Image Segmentation and Feature Extraction to Recognize Online Handwritten Bangla Characters

In this paper, three different feature extraction strategies along with their all possible combinations have been discussed in detail for the recognition of online handwritten Bangla basic characters. Applying a quad-tree based image segmentation approach the target character has been dissected for the extraction of features. Out of these three techniques, one is computing area feature (using composite Simpson's rule) while other two are extracted local (mass distribution and chord length) features. Authors have also investigated optimal depth of the quad-tree (while segmenting an image), at which classifier reveals its best performance. The current experiment has been tested on 10,000 character dataset. Sequential Minimal Optimization (SMO) produces highest recognition accuracy of 98.5 % when all three feature vectors are combined.