Evaluation of Theory based Handwritten Answers through BFO Model for Pixels and Pruned Scale Invariant Character Features

Abstract

Handwritten theory examinations are essential but they come with the complexity of evaluation by subject experts. It requires time consuming efforts from expert evaluators to check answer sheets of many students. Automation has been brought by evaluation of scanned sheets manually by evaluator through an online portal still the effort of evaluator is same and sometimes even more difficult due to problems associated with screen reading. This complete system needs automation such that the system has skills and intelligence equivalent to an expert evaluator to generate scores for the specific subjects. It is essential to recognize handwritings of thousands of candidates, each having unique features. The proposed method has thought of an innovative technique to train the system for every subject and also evaluate the paragraph answer written by the subjects. Segmentation of characters from continuous handwritten text has been done through a novel method inspired from Bacteria Foraging optimization (BFO). The BFO based pixel model spreads bacterial colonies over the text. Healthy colonies are used to identify valid characters while the unhealthy colonies are eliminated. The offspring bacterial colonies produce optimal characters. Final colony arrangements are compared with each other for character recognition through proposed pruned scale invariant features-based method. Handwritten text obtained from standard dataset for more than 50 subjects have been segmented, recognized and scored with optimistic accuracy comparative to the prevalent handwriting recognition methods.