Enhancing The Accuracy Of OCR In RPA Using Meta-Heuristic Technique And Deep Learning

Abstract

Optical Character Recognition (OCR) is recognized as a tool that enables the company to reach scanned paper documents, PDF files, or even digital images taken with a digital camera and convert them into further editable and searchable results. This study focuses on the combination of meta-heuristics with deep learning for improving OCR in the Robotic Process Automation context. OCR, central to capture and extraction of data from documents experiences major hurdles including poor quality images, noise and variation in font. In this study, these limitations are overcome by using Ant Lion Optimization (ALO) and Bat Inspired Algorithm (BIA) for feature selection and model parameters tuning. At the same time, DL architectures such as Bi-LSTM, LSTM and RNN help develop stronger and more effective data pattern recognition and context-sensitive text prediction. The studies show that while the OCR character recognition is raised up to 85.48% and the word reconstruction rate is 69.27%, these results exceed traditional methods. Comparing the proposed framework against other works shows that it is more effective in dealing with noisy, handwritten, and multilingual text. The symbiotic approach has shown vast possibilities for the promotion of intelligent automation not only in the industrial sectors but also to streamline the OCR processes and improve the flexibility to tackle challenging issues closely related to real-life situations.