Year: 2024
Type: article
Abstract: Text recognition from images has many challenges due to differences in the appearance of text such as font, color, size, and background. From such images extracting text accurately is very difficult. Traditional methods attain an imperfect ability to handle such variability and complexity. This study addresses the text recognition problem from images, directing the extraction of text from images with higher accuracy. For accurate and efficient English text recognition, a convolutional neural network (CNN) is proposed. The proposed scheme attains the capability to acquire hierarchical structures from raw pixel values, allowing for the detention of both high-level and low-level information in text images. It also demonstrates promising results on text recognition with an accuracy of 92% compared with Naive Bayes (NB) and Support Vector Machine (SVM). Through evaluation and experiments, CNN captures appropriate information leading to better recognition performance compared to other methods.
DOI: 10.33317/ssurj.644
Authors: Razia Maroof, Irfan Ahmed Usmani, Atruba Feroze
Institutions: Iqra University, Salim Habib University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
Funder: Iqra University
OpenAlex ID: W4405837002