Textract reponse conversion experiments

textract_to_images.py

import boto3
import io
import base64
from PIL import Image # use Pillow

def get_base64_from_layout(image_path, textract_response):
    # Load the original image to get pixel dimensions
    with open(image_path, 'rb') as f:
        img_bytes = f.read()
    image = Image.open(io.BytesIO(img_bytes))
    width, height = image.size

    extracted_images = []
    extracted_tables = []

    # Iterate through blocks to find Tables and Layout Figures
    for block in textract_response.get('Blocks', []):
        block_type = block.get('BlockType')
        
        # Identify Tables or Layout Figures (Images)
        # Note: 'LAYOUT_FIGURE' is returned when using the LAYOUT feature
        if block_type in ['TABLE', 'LAYOUT_FIGURE']:
            bbox = block['Geometry']['BoundingBox']
            
            # Convert normalized coordinates to pixel coordinates
            left = bbox['Left'] * width
            top = bbox['Top'] * height
            right = (bbox['Left'] + bbox['Width']) * width
            bottom = (bbox['Top'] + bbox['Height']) * height
            
            # Crop the image
            crop = image.crop((left, top, right, bottom))
            
            # Convert crop to base64 string
            buffered = io.BytesIO()
            crop.save(buffered, format="PNG")
            img_str = base64.b64encode(buffered.getvalue()).decode('utf-8')
            
            if block_type == 'TABLE':
                extracted_tables.append(img_str)
            else:
                extracted_images.append(img_str)
                
    return {"images": extracted_images, "tables": extracted_tables}

# Example Usage
# response = textract_client.analyze_document(Document={'Bytes': img_bytes}, FeatureTypes=['TABLES', 'LAYOUT'])
# results = get_base64_from_layout("my_document.jpg", response)