Character recognition based on convolutional neural network and Hough transform

Abstract

Handwriting recognition is a delicate task that finds application in several fields, such as the processing of historical documents as well as in the field of security. This study explores the improvement of handwritting recognition by combining a model with the Hough Transform. The integration of the Hough Transform strengthens the robustness of the model, allowing a significant improvement in accuracy. The Hough Transform, adapted to the characteristics of the writing, shows notable efficiency in managing variations in inclination and character size. Evaluated on Kaggle’s A-Z Handwritten Data dataset, the performance of this model reaches an accuracy of 98.53%.