app.py

import gradio as gr
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image, ImageEnhance, ImageOps
import torch
import re
import easyocr
from io import BytesIO
import numpy as np

# Initialize EasyOCR Reader
reader = easyocr.Reader(['en'])

# Load the pre-trained model and processor for fallback
processor = TrOCRProcessor.from_pretrained('microsoft/trocr-small-printed')
model = VisionEncoderDecoderModel.from_pretrained('microsoft/trocr-small-printed')
model.eval()

def enhance_image(image):
    image = ImageEnhance.Contrast(image).enhance(2.0)
    image = ImageEnhance.Sharpness(image).enhance(2.0)
    image = ImageOps.grayscale(image)
    return image


def ocr_with_easyocr(pil_img):
    # Convert PIL to bytes for EasyOCR
    buf = BytesIO()
    pil_img.save(buf, format='PNG')
    return reader.readtext(buf.getvalue(), detail=0)


def ocr_with_trocr(pil_img):
    pixel_values = processor(images=pil_img.convert("RGB"), return_tensors="pt").pixel_values
    with torch.no_grad():
        ids = model.generate(pixel_values)
    return processor.batch_decode(ids, skip_special_tokens=True)[0]


def extract_meter_reading(image):
    try:
        pil = Image.fromarray(image)
        w, h = pil.size
        # Define regions: reading top 40%, serial bottom 40%
        top_region = pil.crop((0, 0, w, int(h * 0.4)))
        bottom_region = pil.crop((0, int(h * 0.5), w, h))
        # Enhance regions
        top_enh = enhance_image(top_region)
        bot_enh = enhance_image(bottom_region)
        # OCR regions
        top_texts = ocr_with_easyocr(top_enh) + [ocr_with_trocr(top_enh)]
        bot_texts = ocr_with_easyocr(bot_enh) + [ocr_with_trocr(bot_enh)]
        top_combined = " ".join(top_texts)
        bot_combined = " ".join(bot_texts)
        # Extract reading: look for digits near kwh
        reading_match = re.search(r"(\d+)\s*(?=kwh|kw h|k w h)", top_combined, re.IGNORECASE)
        if not reading_match:
            # fallback: any 4-7 digit number
            fallback = re.findall(r"\b\d{4,7}\b", top_combined)
            reading = fallback[0] if fallback else "Not Found"
        else:
            reading = reading_match.group(1)
        # Extract serial: longest digit sequence in bottom
        serials = re.findall(r"\d{6,12}", bot_combined)
        serial = max(serials, key=len) if serials else "Not Found"
        return f"Serial Number: {serial}\nMeter Reading: {reading}"
    except Exception as e:
        return f"Error: {str(e)}"

# Gradio app
def main():
    iface = gr.Interface(
        fn=extract_meter_reading,
        inputs=gr.Image(type="numpy", label="Upload or Capture Meter Image"),
        outputs="text",
        title="Meter Reading and Serial Number Extractor",
        description="Upload a meter image; extracts serial number and meter reading using region-based OCR."
    )
    iface.launch()

if __name__ == '__main__':
    main()