Handwritten numeral recognition through superimposition onto printed form

Abstract

Handwriting is, eventually, a variation of the printed forms where the characters are little larger, smaller, angled and deformed than the printed forms. The small changes in handwriting define the parameters of the character to be recognized. Handwritten numeral recognition (HNR) poses significant challenges due to the deformations and other variations. This study proposes a new notion of HNR on the hypothesis that the handwritten numerals are distinct deformations of the printed forms, which leads to easier recognition task with higher accuracy when superimposing handwritten numeral images onto the corresponding printed numeral images. In the proposed HNR, auto-encoder and convolutional auto-encoder have been adapted for the superimposing task that transform HNIs into PNIs, while neural network and convolutional neural network are employed for classification of PNIs. The superimposition method reduces the computational overhead. Moreover, this method employs simple pre-processing withoutfeature extraction whereas the traditional methods employ pre-processing, feature extraction, and recognition with machine learning tools, which add to the computational overhead. The performance of HNRSP has been evaluated for recognizing handwritten numerals of Bengali, Devanagari, and English on benchmark datasets andthe proposed system achieves 99.68%, 99.73%, and 99.62% recognition accuracy for Bengali, Devanagari, and English handwritten numerals, respectively.