app.py

import streamlit as st
import pandas as pd
import numpy as np

import cv2

from PIL import Image

import os

import zipfile
import gdown

from tempfile import TemporaryDirectory
from Utils import Loadlines, decode_predictions, load_model

import shutil
import tempfile
import subprocess

import io

from skimage import filters

# Load the OCR model outside the function to prevent reloading it every time
ocr_model = load_model()




def main():
    
    st.title("منصة تجريبية لمكتب أضاميم")
    st.title("Arabic Manuscript OCR")
    st.warning("يرجى تحميل صورة ذو دقة عالية للحصول على نتيجة أفضل (حوالي 5000*3000)")
    # download_folder()

    # Load sample images
    sample_images_dir = "sample_images"
    sample_images = [os.path.join(sample_images_dir, img) for img in os.listdir(sample_images_dir) if img.endswith(('.png', '.jpg', '.jpeg'))]

    # Display images in a grid and let user select
    col1, col2, col3 = st.columns(3)  # Adjust based on how many images you have; this example assumes 3

    # Normalized size for display
    display_size = (200, 300)

    # Placeholder for selected image
    selected_image_pil = None

    processed_img_path = None

    with col1:
        if st.button("Select Image 1", key="img1"):
            selected_image_pil = display_image(sample_images[0])
        col1.image(resize_image(sample_images[0], display_size), use_column_width=True)

    with col2:
        if st.button("Select Image 2", key="img2"):
            selected_image_pil = display_image(sample_images[1])
        col2.image(resize_image(sample_images[1], display_size), use_column_width=True)

    with col3:
        if st.button("Select Image 3", key="img3"):
            selected_image_pil = display_image(sample_images[2])
        col3.image(resize_image(sample_images[2], display_size), use_column_width=True)


        
    # Option to upload a new image
    uploaded_image = st.file_uploader("Or upload a new image of the Arabic manuscript", type=["jpg", "jpeg", "png"])
    if uploaded_image:
        selected_image_pil = Image.open(uploaded_image)
        st.image(selected_image_pil, caption="Uploaded Image", use_column_width=True)

    
    if selected_image_pil:
        thresh_pil = process_image(selected_image_pil)
        # st.image(thresh_pil, caption="Thresholded Image", use_column_width=True)
    
        # processed_img_path = process_with_yolo(selected_image_pil)
        processed_img_path = process_with_yolo(thresh_pil)
    
        if processed_img_path:
            st.image(processed_img_path, caption="Processed with YOLO", use_column_width=True)
            
            txt_file_path = os.path.join('yolov3/runs/detect/mainlinedection/labels', os.path.basename(processed_img_path).replace(".jpg", ".txt"))
            if os.path.exists(txt_file_path):
                if uploaded_image:
                    original_img_path = uploaded_image  # If you have saved it to a location
                else:
                    original_img_path = next((img for img in sample_images if Image.open(img) == selected_image_pil), None)  # Match selected image with sample images
                display_detected_lines(original_img_path, processed_img_path)
            else:
                st.error("Annotation file (.txt) not found!")

        else:
            st.error("Error displaying the processed image.")

        # display_files_in_directory('/home/user/app/detected_lines/')
            
    

# Function to display files in a directory
# def display_files_in_directory(path="."):
#     if os.path.exists(path):
#         files = os.listdir(path)
#         st.write(f"Files in directory: {path}")
#         for file in files:
#             st.write(file)
#     else:
#         st.write(f"Directory {path} does not exist!")   
  
def resize_image(image_path, size):
    """Function to resize an image"""
    with Image.open(image_path) as img:
        img = img.resize(size)
        return img

def display_image(img):
    """Function to display an image. If img is a path, open it. Otherwise, just display it."""
    if isinstance(img, str):  # img is a file path
        img = Image.open(img)
    
    st.image(img, caption="Selected Image", use_column_width=True)
    return img


# --------------------------------------------

def get_dynamic_kernel(img_height):
    # Set the kernel size to be 1% of the image height
    kernel_height = int(0.001 * img_height)
    
    # Ensure the kernel height is odd
    if kernel_height % 2 == 0:
        kernel_height += 1
    
    # Set minimum and maximum limits
    kernel_height = max(1, kernel_height)  # Minimum limit
    kernel_height = min(11, kernel_height)  # Maximum limit

    st.text(f"img_height : {img_height}")
    st.text(f"kernel_height : {kernel_height}")
    
    return np.ones((kernel_height, 1), np.uint8)
    
def process_image(selected_image):
    # Convert PIL image to OpenCV format
    opencv_image = np.array(selected_image)

    # Check if the image is grayscale or RGB
    if len(opencv_image.shape) == 3 and opencv_image.shape[2] == 3:  # RGB image
        gray_image = cv2.cvtColor(opencv_image, cv2.COLOR_RGB2GRAY)
    else:  # Image is already grayscale
        gray_image = opencv_image

    # Ensure the image is 8-bit grayscale
    if gray_image.dtype != np.uint8:
        gray_image = (gray_image * 255).astype(np.uint8)

    # Apply adaptive thresholding

    # Let's assume `gray_image` is your grayscale image array
    height, width = gray_image.shape
    
    # Example: setting block size to 1/30th of the average image dimension, making sure it's odd.
    block_size = ((height + width) // 2) // 30
    block_size = block_size + 1 if block_size % 2 == 0 else block_size
    
    # Example: setting offset to a small fraction of the global mean intensity.
    offset = np.mean(gray_image) * 0.05
    
    adaptive_threshold = filters.threshold_local(gray_image, block_size, offset=offset, method='mean')
    
    binary_adaptive = gray_image > adaptive_threshold

    thresh = binary_adaptive
        
    # Convert thresholded image back to PIL format to display in Streamlit
    thresh_pil = Image.fromarray(thresh)

    return thresh_pil




def get_detected_boxes(txt_path, img_width, img_height):
    with open(txt_path, 'r') as f:
        lines = f.readlines()
        boxes = []
        for line in lines:
            parts = list(map(float, line.strip().split()))
            # Assuming the format is: class x_center y_center width height confidence
            x_center, y_center, width, height = parts[1:5]
            
            # Convert to pixel values
            x_center *= img_width
            y_center *= img_height
            width *= img_width
            height *= img_height

            boxes.append([x_center, y_center, width, height])

    return boxes
    
    



# ---------------------------------------------------------------------------------

def download_ultralytics_yolov3_folder():
    st.text("Downloading Ultralytics YOLOv3 folder from Google Drive. This may take a while...")

    # url = 'https://drive.google.com/file/d/1n6YcqHl5Y2xRpWw7DQPrZ2FoAqfcf12I/view?usp=share_link'
    url = 'https://drive.google.com/uc?id=1n6YcqHl5Y2xRpWw7DQPrZ2FoAqfcf12I'

    output = 'ultralytics_yolov3.zip'
    gdown.download(url, output, quiet=False)

    # Extracting the zip file
    with zipfile.ZipFile(output, 'r') as zip_ref:
        zip_ref.extractall('.')
        st.text("Folder extraction complete!")
        
    os.remove(output)  # Optional: remove the downloaded zip file after extraction.
    st.text("Download and extraction completed!")


def process_with_yolo(img_pil):


    with st.spinner('Downloading YOLOv3 folder...'):
        download_ultralytics_yolov3_folder()

    # display_files_in_directory('/home/user/app/')
    # Save the image to a temporary file
    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".jpg")
    img_pil.save(temp_file.name)

    if os.path.exists('yolov3/runs/detect/mainlinedection'):
        shutil.rmtree('yolov3/runs/detect/mainlinedection')

    cmd = [
    'python', 'yolov3/detect.py',
    '--source', temp_file.name,  # use the temp file path here
    '--weights', 'yolov3/runs/train/mainline/weights/best.pt',
    '--save-txt', 
    '--save-conf',
    '--imgsz', '672',
    '--name', 'mainlinedection'
    ]
    process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
    
    stdout, stderr = process.communicate()
    if process.returncode != 0:
        st.error(f"YOLOv3 command failed with code {process.returncode}.")

    


    # After processing the image, at the end of the function...
    output_path = os.path.join('yolov3/runs/detect/mainlinedection', os.path.basename(temp_file.name))
    if os.path.exists(output_path):
        return output_path
    else:
        st.error("Processed image not found!")
        return None
    
    # Optional: You can print the output to Streamlit, though it might be extensive.
    st.write(stdout.decode())
    if stderr:
        st.error(stderr.decode())

    # Optional: Close and delete the temporary file
    temp_file.close()
    os.unlink(temp_file.name)




def display_detected_lines(original_path, output_path):
    # Derive the txt_path from the output_path
    txt_path = os.path.join('yolov3/runs/detect/mainlinedection/labels', os.path.basename(output_path).replace(".jpg", ".txt"))

    if os.path.exists(txt_path):
        # Load both original and thresholded images
        original_image = Image.open(original_path)
        thresholded_image = process_image(original_image)
        boxes = get_detected_boxes(txt_path, original_image.width, original_image.height)

        if not boxes:
            st.warning("No lines detected by YOLOv3.")
            return

        # Create a temporary directory to store the detected lines
        with TemporaryDirectory() as temp_dir:
            # detected_line_paths = []  # For storing paths of the thresholded line images for OCR
            original_line_paths = []  # For storing paths of the original line images for display

            for index, box in enumerate(boxes):
                x_center, y_center, width, height = box
                x_min = int(x_center - (width / 2))
                y_min = int(y_center - (height / 2))
                x_max = int(x_center + (width / 2))
                y_max = int(y_center + (height / 2))

                # Crop the ORIGINAL image and save
                original_line = original_image.crop((x_min, y_min, x_max, y_max))
                original_line_path = os.path.join(temp_dir, f"original_line_{index}.jpg")
                original_line.save(original_line_path)
                original_line_paths.append(original_line_path)

                # # Crop the THRESHOLDED image and save for OCR
                # extracted_line = thresholded_image.crop((x_min, y_min, x_max, y_max))
                # detected_line_path = os.path.join(temp_dir, f"detected_line_{index}.jpg")
                # extracted_line.save(detected_line_path)
                # detected_line_paths.append(detected_line_path)

            # Perform OCR on thresholded lines
            recognized_texts = perform_ocr_on_detected_lines(original_line_paths)

            # Display the results
            for original_img_path, text in zip(original_line_paths, recognized_texts):
                st.image(original_img_path, use_column_width=True)
                st.markdown(
                    f"<p style='font-size: 18px; font-weight: bold;'>{text}</p>",
                    unsafe_allow_html=True
                )
                # Add a small break for better spacing
                st.markdown("<br>", unsafe_allow_html=True)

    else:
        st.error("Annotation file (.txt) not found!")



def perform_ocr_on_detected_lines(detected_line_paths):
    """
    Performs OCR on the provided list of detected line image paths.
    
    Args:
    - detected_line_paths: List of paths to the detected line images.

    Returns:
    - A list of recognized text for each image.
    """

    # Load the saved detected lines for OCR processing
    test_dataset = Loadlines(detected_line_paths)
    
    prediction_texts = []
    for batch in test_dataset:
        preds = ocr_model(batch) 
        pred_texts = decode_predictions(preds)
        # st.text(f"Decoded OCR Results : {pred_texts}")
        prediction_texts.extend(pred_texts)
        
    return prediction_texts

    
if __name__ == "__main__":
    main()