Printed Text Scanner (PyTesseract + PyQt5)

Lightweight GUI to load images or capture webcam frames, draw a rectangular ROI, and extract printed text using Tesseract OCR. The app is implemented in a single script, test.py, using PyQt5 for the interface and OpenCV + pytesseract for image processing and OCR.

Key features (as implemented in test.py):

• Load images (PNG/JPG/JPEG/BMP) using a file dialog. Uses cv2.imdecode + numpy.fromfile to support long/Unicode paths on Windows.
• Webcam support: start/stop camera with cv2.VideoCapture and a QTimer to show live frames.
• Drawable ROI: click-and-drag on the image to draw a rectangular ROI. The ImageLabel class converts between display and image coordinates so the ROI maps correctly to the underlying image.
• OCR with preprocessing: selected region (or full image) is converted to grayscale, filtered with cv2.bilateralFilter, and binarized with cv2.adaptiveThreshold before calling Tesseract.
• Bounding boxes & overlays: high-confidence OCR words are drawn as rectangles with text overlays on the image.
• Text output: full OCR text is placed in the right-side text panel.
• Save overlay: save the current overlay image to disk (PNG) using cv2.imencode and tofile.

Requirements:

• Python 3.8+
• A Tesseract OCR binary installed (and available on PATH, or set manually)
• Python packages:
  • PyQt5
  • opencv-python
  • pillow
  • pytesseract
  • numpy

Install dependencies (Windows PowerShell):

python -m pip install --upgrade pip
pip install PyQt5 opencv-python pillow pytesseract numpy

Or create a requirements.txt and run:

pip install -r requirements.txt

Tesseract on Windows:

• Install Tesseract (e.g. UB Mannheim build) and ensure tesseract.exe is on your PATH.
• If Tesseract isn't on PATH, set it in test.py before calls to pytesseract:

import pytesseract
pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"

Run the app:

python test.py

Usage flow in the GUI:

• Click Load Image or Start Camera to get an image into the viewer.
• Draw a rectangle by dragging with the left mouse button to define an ROI (optional).
• Click Capture Frame to mark the current frame (shows a confirmation message).
• Click Run OCR to run preprocessing + Tesseract on the ROI or full image.
• OCR results appear in the right-side text area; high-confidence words are overlaid on the image.
• Click Save Overlay to export the displayed image with overlays as a PNG.

Implementation / behavior notes (from the code):

• ROI mapping: ImageLabel computes scaling between the pixmap and the label size. The ROI saved in image coordinates so overlays and OCR operate on the correct pixels.
• Preprocessing pipeline (in run_ocr): grayscale -> bilateral filter -> adaptive Gaussian thresholding. Output of preprocessing is passed to Tesseract via Pillow image conversion.
• OCR box filtering: the OCR output uses pytesseract.image_to_data and filters words by confidence > 40 (integer-conf). Change this value in test.py to relax or tighten filtering.
• Saving images: display_image is kept as RGB; when saving the code converts to BGR and writes via cv2.imencode(...).tofile(path) to support Unicode/long paths on Windows.

Troubleshooting & tips:

• If images fail to load, confirm the path has read permissions and contains supported formats. Using non-ASCII characters in file paths is supported by the current cv2.imdecode approach but may still cause issues in some environments.
• If OCR returns nothing or poor text, try:
  • Increasing image contrast or experimenting with different preprocessing (e.g., denoising, resizing).
  • Lowering the confidence threshold in the code to see more raw detections.
  • Installing language packs for Tesseract if you expect non-English text.

Where the code lives:

• Main GUI and OCR logic: test.py

Next steps I can help with:

• Add a requirements.txt and a small run.ps1 helper script.
• Update test.py to auto-detect Tesseract and set pytesseract.pytesseract.tesseract_cmd when missing (with a helpful UI message).
• Add a command-line mode to run OCR headless on images or directories.

If you want any of those, tell me which and I will implement it.