Attention-Based Deep Learning Algorithm in Natural Language Processing for Optical Character Recognition

Abstract

Optical Character Recognition (OCR) is commonly referred as text recognition which poses a substantial issue in the computer vision tasks. Conventional optical character recognition systems frequently suffer in handwritten document recognition. To solve this, Deep Learning (DL) models have emerged as a powerful and advanced solution for character recognition. The present research offers a unique CNN-RNN model with an Attention Mechanism (CNN-RNN-AM) for English image character recognition. The process comprised many important phases, beginning with image collection from a user-defined dataset, then image pre-processesing includes grayscale conversion and noise reduction. For effective character recognition, the proposed approach integrates the segmentation process at multiple levels, including line segmentation, word segmentation, and character segmentation. Finally, the CNN-RNN with an attention mechanism is deployed for character recognition. The experimental findings demonstrated the remarkable efficacy of the suggested CNN-RNN-AM model. It outperformed other compared models by attaining an excellent character recognition accuracy of 99.89%.