Poposed a Model to Improve Handwritten Number Recognition Using Convolutional Neural Networks

Considering the prevalence of handwritten documents in human transactions, optical character recognition (OCR) of documents is extremely valuable. OCR is a science that enables the translation of various types of documents or images into analyzable, editable, and searchable data. In this regard, the aim of this research is to recognize handwritten Persian digits using convolutional neural networks. To evaluate the performance of deep learning models, the numbers in the IFHCDB dataset were used. According to the studies, this research has selected four of the top deep learning architectures to advance its goals, namely VGG-۱۹, ResNet-۱۰۱, MobileNet, and ConvNext. In the first phase of this research, the models were initially trained without preprocessing. In the second phase, by using preprocessing and changing the architecture of the models, a proposed method was presented to improve the accuracy and performance of the models. Experimental results have shown that preprocessing and the proposed architecture have led to a significant improvement in the performance of the models. In addition, the proposed method has been able to achieve the highest results among the related works in number recognition using convolutional neural networks on the IFHCDB dataset in two models, ConvNext with an accuracy of ۹۹.۱۶ and VGG-۱۹ with an accuracy of ۹۹.۰۸.

کلیدواژه ها:

Farsi OCR ، Persian OCR ، Convolution Neural Network ، CNN