Title of Submission
Electrical and Computer Engineering, PhD
Major Advisor Name
Elisa Barney Smith
Type of Submission
This research is about offline Bangla handwriting text recognition. Here we introduce a publicly accessible dataset, as well as a basic character recognition scheme. The dataset contains pages with a 104 word essay and a collection of 84 isolated alpha-numeric characters. All the components in the pages are tagged with the associated ground truth information. The character recognition scheme presented here uses zonal pixel counts, structural strokes and bag of features modeled with grid points using U-SURF descriptor as features. The maximum classification accuracy we obtain is 96.8% using an SVM classifier with a cubic kernel.