Invoice OCR Fraud Detection System

Developed a system to extract text from invoice documents using OCR (Optical Character Recognition) techniques.

Implemented pre-processing steps (noise removal, thresholding, skew correction) to improve text extraction accuracy.

Used Tesseract OCR / EasyOCR to automatically read fields like invoice number, date, vendor details, and total amount.

Applied fraud-detection rules and machine learning to identify anomalies such as mismatched amounts, duplicate invoice numbers, fake vendor names, and altered documents.

Integrated NLP and pattern-matching to validate critical fields (like GSTIN format, invoice ID structure).

Built a structured pipeline converting raw invoice images into usable tabular data for backend verification.

Designed visual dashboard/alerts to flag potentially fraudulent invoices for human verification.

System improves financial security by reducing manual efforts, preventing billing fraud, and ensuring audit transparency.

Evaluated performance using accuracy and false-positive rate; optimized model with real-world invoice datasets.

Technology stack may include: Python, OpenCV, Tesseract, Scikit-learn / ML, Flask/Streamlit, SQL