A free Python-based document and image processing platform.

• Image and scanned-document preprocessing
• Automatic deskewing
• Noise and shadow reduction
• PDF page rendering
• English and Bengali OCR
• OCR confidence scoring
• Word-level bounding boxes
• TXT, CSV and JSON export
• Human-review routing

Input image or PDF → preprocessing → clean and binary images → multilingual OCR → confidence analysis → approved output or review queue

• Windows 11
• Python 3.11
• Tesseract OCR 5
• English and Bengali Tesseract language models

python -m pip install -r requirements.txt

python src/preprocessing/preprocess_documents.py --input "PATH_TO_IMAGE_OR_PDF"

python src/ocr/ocr_documents.py --input "PROCESSED_FOLDER" --lang "eng"

python src/ocr/ocr_documents.py --input "PROCESSED_FOLDER" --lang "ben+eng" --review-threshold 90

Do not commit client documents, personal information, OCR outputs, datasets, models, logs, credentials or company-confidential materials.

This is an early MVP. Bengali OCR requires further improvement, confidence calibration and human quality control.

Three model modes are supported:

• fast: faster CPU processing and the default
• best: larger, slower Tesseract models
• active: uses the files in models/tesseract/tessdata

Example:

powershell.exe -ExecutionPolicy Bypass -File ".\run_iimi.ps1" `
    -InputPath "PATH_TO_IMAGE_OR_PDF" `
    -Language "ben+eng" `
    -Model "fast" `
    -ReviewThreshold 90

Model quality must be checked against verified ground-truth text. A higher OCR confidence score does not by itself prove higher accuracy.