Integration of the Hough transform method into the layers of convolutional neural network for character recognition

Abstract

Handwritten character recognition is a complex issue with diverse applications, particularly in the fields of historical document archiving and security. This research proposes an innovative approach to optimize this task by integrating the Hough Transform with in the Layers of a Convolutional Neural Network. This hybridization increases the robustness of the model, leading to a significant improvement in classification performance. Specifically adapted to the morphological particularities of writing, the Hough Transform demonstrates particular effectiveness in taking into account the angular and dimensional variations of characters. Experiments conducted on Kaggle’s A-Z handwritten database attest to the effectiveness of this method, with an accuracy rate reaching 98.75%.