Abstract

This paper presents the development and deployment of the DocVerify App, an AI-based solution designed to enhance document authentication processes. By utilizing OCR and similarity algorithms, the app extracts and compares text from images and PDFs to detect inconsistencies. The project aims to provide businesses and institutions with a robust mechanism to prevent forgery, streamline verification, and maintain high data accuracy.

Introduction

Forgery and document manipulation have become pervasive issues across sectors such as finance, legal, and healthcare. Traditional verification methods often fall short in detecting subtle changes in official records. This project introduces an AI-driven document verifier that uses OCR and cosine similarity algorithms to automate document validation, minimizing human error and increasing efficiency.

The app's goal is to simplify the verification process by allowing users to upload documents for comparison with original templates, identifying mismatches and alerting users to inconsistencies.

Methodology

System Architecture
The DocVerify App follows a modular design consisting of the following components:

• Frontend: Developed with Streamlit for user interaction and file uploads.
• Backend: Uses Python libraries like OpenCV and Tesseract OCR for image processing and text extraction.
• NLP Engine: Employs cosine similarity and Levenshtein distance algorithms to compare extracted text.

OCR and Text Extraction
Tesseract OCR is used to extract textual data from scanned images and PDFs. The extracted text is preprocessed by removing noise, special characters, and formatting inconsistencies.

Text Similarity Analysis
Text comparison is performed using cosine similarity and vectorization techniques. This helps the app detect minute variations between original and subsequent documents.

Experiments

Dataset and Document Types
The experiment was conducted using a dataset comprising:

• 100 original documents (contracts, certificates, invoices)
• 100 altered documents with varying levels of manipulation
• Documents were scanned at resolutions of 300 DPI to ensure OCR accuracy.

Testing Environment

• Hardware: Intel i7, 16GB RAM
• Software: Python 3.9, Streamlit, Tesseract 5.4.0, OpenCV 4.7
• OS: Windows 11

Results

The Document Verifier App demonstrated the following performance:

• Accuracy: 94% in detecting discrepancies
• False Positives: 3%
• False Negatives: 5%
• Processing Time: 1.5 seconds per document

The app successfully identified tampered documents

Conclusion

The DocVerify App proves to be an effective solution for automating document authentication. By integrating OCR and similarity analysis, the app offers a high-accuracy tool for detecting document forgery.

Future improvements will focus on enhancing multilingual support, integrating AI-driven anomaly detection, and expanding the app’s capabilities to handle larger document volumes.

This project highlights the potential of AI and computer vision in strengthening document verification processes, contributing to fraud prevention across industries.