Das, Abhijit, Suwanwiwat, Hemmaphan, and Pal, Umapada (2023) A novel multi-task learning technique for offline handwritten short answer spotting and recognition. Multimedia Tools and Applications. (In Press)

PDF (Publisher Accepted Version) - Published Version Restricted to Repository staff only

Abstract

Off-line examination is still being used in many parts of the world as it is a more economical way of conducting exams when compared to computer-based ones. Automatically and accurately assessing these handwritten exam papers poses a complex challenge, as high accuracy rates as possible are always desirable. Factors such as the attributes of the handwritten images, the presence of numerous classes, challenges related toword boundaries in languages such as Arabic, and the significant intra-class variation in handwritten forms contribute to the enduring complexity of word recognition and word spotting tasks. In order to address the problems, this research proposed a novel joint learning technique for word spotting and word recognition in a multi-task learning setting. A multi-task convolution neural network was employed to materialise the proposed concept. The word spotting task was dealt as a regression task and the other task was word recognition. The typical Faster-RCNN backbone was employed with the Region of Interest (RoI) pooling layer, which was then followed by two consecutive fully connected layers for the word spotting and recognition task. The experimental results are encouraging and demonstrate that the proposed research achieved a significant enhancement in the accuracy of short-answer assessment systems. As a result, the proposed technique can be implemented in short-answer assessment systems to improve both their efficiency and accuracy.

Item ID:	81171
Item Type:	Article (Research - C1)
ISSN:	1573-7721
Keywords:	Word spotting, Short answer assessment, Multi-task learning technique
Copyright Information:	© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023
Date Deposited:	20 Nov 2023 23:39
FoR Codes:	46 INFORMATION AND COMPUTING SCIENCES > 4603 Computer vision and multimedia computation > 460308 Pattern recognition @ 100%
SEO Codes:	22 INFORMATION AND COMMUNICATION SERVICES > 2204 Information systems, technologies and services > 220403 Artificial intelligence @ 100%
Downloads:	Total: 1
	More Statistics