Scopus Indexed Publications

Paper Details


Title
MathNET: Using CNN Bangla Handwritten Digit, Mathematical Symbols, and Trigonometric Function Recognition
Author
Shifat Nayme Shuvo, Fuad Hasan, Sheikh Abujar, Syed Akhter Hossain,
Email
shifat15-9836@diu.edu.bd
Abstract

Scientific methods are mostly based on mathematical solutions. Optical character recognition (OCR) is one of the most required solution for digitalizing and/or processing the handwritten documents into any digital form. Therefore, the movement toward the use of digital documents in the scientific community has significantly increased. Scientific documents and more precise mathematical documents are sometimes initially drafted in handwritten form. In this scope, to let understand the machine about those mathematical equations, so in this paper, a research is done on scripted mathematical symbol recognition model, which was built of 32,400 images dataset. So many research in English handwritten OCR has already achieved very good accuracy, but for Bengali language handwritten OCR, there are a minimum number of research that is done till now. This dataset contains samples of handwritten mathematical symbols and especially Bengali numerical digits. In detail, this dataset captures—Bangla numerical digits, operators of algebra, set-symbols, limit, calculus, symbols of comparison, delimiters, etc. Thus, MathNET is presented as a model that helps to recognize 10 handwritten Bangla digits and 44 other handwritten mathematical symbols. The planned model training accuracy is 96.01% and validation accuracy with test data is 96.50%, which is very good precision for identification of mathematical symbols.

Keywords
Bangla handwritten digits Data processing Pattern recognition CNN Mathematical symbols Trigonometrical function OCR
Journal or Conference Name
Soft Computing Techniques and Applications. Advances in Intelligent Systems and Computing
Publication Year
2021
Indexing
scopus