app.py

import os
from PIL import Image, ImageOps, ImageChops
import io
import fitz  # PyMuPDF
from docx import Document
from rembg import remove
import gradio as gr
from hezar.models import Model
from ultralytics import YOLO
import json

# ایجاد دایرکتوری‌های لازم
os.makedirs("static", exist_ok=True)
os.makedirs("output_images", exist_ok=True)

def trim_whitespace(image):
    gray_image = ImageOps.grayscale(image)
    inverted_image = ImageChops.invert(gray_image)
    bbox = inverted_image.getbbox()
    trimmed_image = image.crop(bbox)
    return trimmed_image

def convert_pdf_to_images(pdf_path, zoom=2):
    pdf_document = fitz.open(pdf_path)
    images = []
    for page_num in range(len(pdf_document)):
        page = pdf_document.load_page(page_num)
        matrix = fitz.Matrix(zoom, zoom)
        pix = page.get_pixmap(matrix=matrix)
        image = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)
        trimmed_image = trim_whitespace(image)
        images.append(trimmed_image)
    return images

def convert_docx_to_jpeg(docx_bytes):
    document = Document(BytesIO(docx_bytes))
    images = []
    for rel in document.part.rels.values():
        if "image" in rel.target_ref:
            image_stream = rel.target_part.blob
            image = Image.open(BytesIO(image_stream))
            jpeg_image = BytesIO()
            image.convert('RGB').save(jpeg_image, format="JPEG")
            jpeg_image.seek(0)
            images.append(Image.open(jpeg_image))
    return images

def remove_background_from_image(image):
    return remove(image)

def process_file(input_file):
    file_extension = os.path.splitext(input_file.name)[1].lower()
    images = []

    if file_extension in ['.png', '.jpeg', '.jpg', '.bmp', '.gif']:
        image = Image.open(input_file)
        image = image.convert('RGB')
        output_image = remove_background_from_image(image)
        images.append(output_image)
    elif file_extension == '.pdf':
        images = convert_pdf_to_images(input_file.name)
        images = [remove_background_from_image(image) for image in images]
    elif file_extension in ['.docx', '.doc']:
        images = convert_docx_to_jpeg(input_file.name)
        images = [remove_background_from_image(image) for image in images]
    else:
        return "File format not supported."

    input_folder = 'output_images'
    for i, img in enumerate(images):
        img.save(os.path.join(input_folder, f'image_{i}.jpg'))

    return images

def run_detection_and_ocr():
    # Load models
    ocr_model = Model.load('hezarai/crnn-fa-printed-96-long')
    yolo_model = YOLO("/content/drive/MyDrive/train3/weights/best.pt")
    
    input_folder = 'output_images'
    yolo_model.predict(input_folder, save=True, imgsz=320, conf=0.5, save_crop=True)
    
    output_folder = '/content/runs/detect/predict'
    results = []
    
    for filename in os.listdir(input_folder):
        if filename.endswith('.JPEG') or filename.endswith('.jpg'):
            image_path = os.path.join(input_folder, filename)
            crop_folder = os.path.join(output_folder, 'crops')
            crops = []
            for crop_label in os.listdir(crop_folder):
                crop_label_folder = os.path.join(crop_folder, crop_label)
                if os.path.isdir(crop_label_folder):
                    for crop_filename in os.listdir(crop_label_folder):
                        crop_image_path = os.path.join(crop_label_folder, crop_filename)
                        text_prediction = predict_text(ocr_model, crop_image_path)
                        crops.append({
                            'crop_image_path': crop_image_path,
                            'text_prediction': text_prediction,
                            'class_label': crop_label
                        })
            results.append({
                'image': filename,
                'crops': crops
            })

    output_json_path = 'output.json'
    with open(output_json_path, 'w', encoding='utf-8') as f:
        json.dump(results, f, ensure_ascii=False, indent=4)
    
    return output_json_path

def predict_text(model, image_path):
    try:
        image = Image.open(image_path)
        image = image.resize((320, 320))
        output = model.predict(image)
        if isinstance(output, list):
            return ' '.join([item['text'] for item in output])
        return str(output)
    except FileNotFoundError:
        return "N/A"

def gradio_interface(input_file):
    process_file(input_file)
    json_output = run_detection_and_ocr()
    with open(json_output, 'r', encoding='utf-8') as f:
        return json.load(f)

iface = gr.Interface(
    fn=gradio_interface,
    inputs=gr.File(label="Upload Word, PDF, or Image"),
    outputs=gr.JSON(label="JSON Output"),
    title="Document to JSON Converter with Background Removal"
)

if __name__ == "__main__":
    iface.launch()