Function detect_text

StegoScan.py:738–757

Runs OCR (TrOCR) and ensures only meaningful text is considered.

(image)

def detect_text(image):
    """Runs OCR (TrOCR) and ensures only meaningful text is considered."""
    # Preprocess image for better OCR accuracy
    preprocessed_img = enhance_image_for_ocr(image)

    # Convert to PIL Image for TrOCR processing
    pil_image = Image.fromarray(preprocessed_img).convert("RGB")

    # Get pixel values for model input
    pixel_values = processor_handwritten(images=pil_image, return_tensors="pt").pixel_values.to(device)

    # Perform OCR prediction
    generated_ids = model_handwritten.generate(pixel_values)
    transcription = processor_handwritten.decode(generated_ids[0], skip_special_tokens=True).strip()

    # Apply filtering: Ignore results if text length is too short or contains mostly non-alphanumeric characters
    if len(transcription) < 3 or sum(c.isalnum() for c in transcription) / len(transcription) < 0.5:
        return False, ""  # Ignore if it's too short or mostly symbols

    return True, transcription


# I think the above line is making it to where there is going to save all the file variations