Payment-Verification-OCR-2025

Automated payment verification system using OCR and AI-powered object detection to extract and validate transaction IDs from payment screenshots.

Features

• Extracts transaction IDs from payment screenshots using YOLOv12 for cropping and pytesseract for OCR.
• Supports popular platforms: PhonePe, Google Pay, Paytm, Amazon Pay.
• Accepts transaction reports in CSV, Excel, and PDF formats.
• Automatically matches extracted transaction IDs with backend transaction logs.
• Detects duplicate and registration duplicate transaction IDs.
• Marks registrations as "Verified", "Not Verified", "Duplicate", "Registration Duplicate", "No ID extracted", or "Amount mismatch".
• Streamlit Web UI for easy file upload, column configuration, and result download.
• Fallback mechanism: Uses registration form transaction IDs when OCR fails.
• Docker support for deployment.

Tech Stack

• Python — Core language
• YOLOv12 — Object detection for cropping transaction ID regions
• pytesseract — OCR text extraction
• Streamlit — Web UI
• Pandas — Data processing
• OpenCV — Image processing
• pdfplumber / PyPDF2 — PDF transaction report parsing
• Docker — Containerized deployment

Project Structure

Payment-Verification-OCR-2025/
├── app.py                 # Streamlit web UI (main entry point)
├── extraction.py          # OCR extraction module (YOLO + pytesseract)
├── ID_verify.py           # Transaction ID verification module
├── pipeline.py            # CLI runner for batch processing
├── model.pt               # YOLOv12 model weights
├── requirements.txt       # Python dependencies
├── .gitignore             # Git ignore rules
└── README.md              # This file

How It Works

1. Upload the registration CSV/Excel file (with screenshot URLs) via the web UI.
2. Upload one or more transaction report files (CSV, Excel, or PDF).
3. Configure which columns contain transaction IDs, RRNs, and amounts.
4. The system:
   • Downloads each screenshot from the URLs.
   • Uses YOLOv12 to detect and crop the transaction ID region.
   • Applies pytesseract OCR to extract the transaction ID.
   • Falls back to the registration form ID if OCR fails.
   • Matches extracted IDs against the uploaded transaction reports.
   • Checks for duplicates and amount mismatches.
5. Download the final verification report as CSV.

Quick Start

Web UI (Streamlit)

pip install -r requirements.txt
streamlit run app.py

CLI (Batch Processing)

# Place input.csv (with "screenshot" column) in the project root
python pipeline.py

Supported Receipt Types

Platform	ID Format
PhonePe	UTR numbers starting with T + 21 digits
Google Pay	Transaction IDs like AXIS1234567890
Paytm	12–15 digit numeric reference numbers
Amazon Pay	Bank Reference ID (alphanumeric)

System Requirements

• Python 3.11+
• Tesseract OCR installed on the system
  • macOS: brew install tesseract
  • Ubuntu/Debian: sudo apt-get install tesseract-ocr
  • Windows: Download installer