An OCR-Based Intelligent System for Automated Invoice Data Extraction
J. Swapna Priya,
Shaik Anwar,
Shaik Farukh,
Sayampu Vamsi,
Shaik Nadeem
Accurate and efficient extraction of invoice data is an essential requirement in modern business and
financial operations, where organizations process large numbers of in- voices every day. Traditional
manual invoice handling is time- consuming, labor-intensive, and highly error-prone. Conventional
Optical Character Recognition (OCR) systems offer partial au- tomation by converting document
images into machine-readable text, but they often fail to extract critical invoice fields reliably because
of variations in layout, font style, image quality, scanning distortions, and background noise. These
limitations lead to unstructured outputs and require extensive human verification. This paper presents
an OCR-based intelligent system for auto- mated invoice data extraction by integrating YOLO-based
object detection with selective OCR. Instead of applying OCR to the entire invoice image, the proposed
system first detects important invoice fields such as invoice number, date, vendor details, and total
amount. OCR is then applied only to the detected regions of interest, which reduces noise and improves
recognition accuracy. The extracted text is further refined using rule-based validation and pattern
matching to generate structured outputs in CSV and Excel-compatible form. The reformatted study
reports an extraction accuracy of approximately 92–95% and a processing speed improvement of nearly
35–40% over a conventional OCR- only approach, demonstrating the practical value of combining
region localization and targeted text recognition for invoice automation.
