BoZFEx – A New Feature Extraction Based on Bounded-Zone Method for Handwriting Character Recognition

The selection of a suitable set of features to represent input samples is one of the most influential elements in a successful optical character recognition, OCR pipeline. The feature extraction phase plays a vital role in extracting the appropriate features that yield minimum classification error, especially in cursive or fibrous handwriting. This study proposed a new feature extraction; BoZFEx - Bounded-Zone Feature Extraction, which focuses on the character recognition of children's handwriting. Minnesota handwriting assessment (MHA) - an established handwriting assessment was used in data collection that uses English alphabets. A total of 4841 characters were collected from 90 children with two font sizes. The proposed feature extraction method is integrated into the optical character recognition (OCR) pipeline. Among font size 36, there were 4 vowel letters, a, e, o, and u showing the consistency of the number of pixels in zone 11 and zone l2. This method was compared with the established character recognition method using a machine learning algorithm, i.e., a multi-input convolutional neural network and achieved promising results with 0.78 accuracy.