app.py

# Import needed library
from PIL import Image
import gradio as gr
import torch
import requests
import re
from transformers import pipeline, GenerationConfig ,BlipProcessor, BlipForConditionalGeneration, TrOCRProcessor, VisionEncoderDecoderModel

# load image examples
img_urls_1 = ['https://i.pinimg.com/564x/f7/f5/bd/f7f5bd929e05a852ff423e6e02deea54.jpg', 'https://i.pinimg.com/564x/b4/29/69/b4296962cb76a72354a718109835caa3.jpg',
        'https://i.pinimg.com/564x/f2/68/8e/f2688eccd6dd60fdad89ef78950b9ead.jpg']
for idx1, url1 in enumerate(img_urls_1):
  image = Image.open(requests.get(url1, stream=True).raw)
  image.save(f"image_{idx1}.png")

# load image examples
img_urls_2 = ['https://i.pinimg.com/564x/14/b0/07/14b0075ccd5ea35f7deffc9e5bd6de30.jpg', 'https://newsimg.bbc.co.uk/media/images/45510000/jpg/_45510184_the_writings_466_180.jpg',
        'https://cdn.shopify.com/s/files/1/0047/1524/9737/files/Cetaphil_Face_Wash_Ingredients_Optimized.png?v=1680923920', 'https://github.com/kawther12h/Image_Captioning-and-Text_Recognition/blob/main/handText22.jpg?raw=true','https://github.com/kawther12h/Image_Captioning-and-Text_Recognition/blob/main/handText11.jpg?raw=true']
for idx2, url2 in enumerate(img_urls_2):
  image = Image.open(requests.get(url2, stream=True).raw)
  image.save(f"tx_image_{idx2}.png")

# Load Blip model and processor for captioning
processor_blip = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
model_blip = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")

# Load marefa model for translation (English to Arabic)
translate = pipeline("translation",model="marefa-nlp/marefa-mt-en-ar")

def caption_and_translate(img, min_len, max_len):
    # Generate English caption
    raw_image = Image.open(img).convert('RGB')
    inputs_blip = processor_blip(raw_image, return_tensors="pt")
    # Create GenerationConfig with max_length
    generation_config = GenerationConfig(max_length=max_len)

    out_blip = model_blip.generate(
        **inputs_blip,
        min_length=min_len,
        max_length=max_len,
        config=generation_config  # Pass the generation_config object here
    )
    #out_blip = model_blip.generate(**inputs_blip, min_length=5, max_length=50)
    english_caption = processor_blip.decode(out_blip[0], skip_special_tokens=True)

    # Translate caption from English to Arabic
    arabic_caption = translate(english_caption)
    arabic_caption = arabic_caption[0]['translation_text']
    translated_caption = f'<div dir="rtl">{arabic_caption}</div>'

    # Return both captions
    return english_caption, translated_caption


# Gradio interface with multiple outputs
img_cap_en_ar = gr.Interface(
    fn=caption_and_translate,
    inputs=[gr.Image(type='filepath', label='Image'),
            gr.Slider(label='Minimum Length', minimum=1, maximum=500, value=30),
            gr.Slider(label='Maximum Length', minimum=1, maximum=500, value=100)],
    outputs=[gr.Textbox(label='English Caption'),
             gr.HTML(label='Arabic Caption')],
    title='Image Captioning | وصف الصورة',
    description="Upload an image to generate an English & Arabic caption | قم برفع صورة وأرسلها ليظهر لك وصف للصورة",
    examples =[["image_2.png"]]
)


# Load the model
text_rec = pipeline("image-to-text", model="jinhybr/OCR-Donut-CORD")

# Load MarianMT model for translation (English to Arabic)
translate = pipeline("translation",model="marefa-nlp/marefa-mt-en-ar")

# Function to process the image and extract text
def extract_text(image):
    # Pass the image to the pipeline
    result = text_rec(image)

    # Extract the plain text and remove tags
    text = result[0]['generated_text']
    text = re.sub(r'<[^>]*>', '', text)  # Remove all HTML tags

    # Translate extracted text from English to Arabic
    arabic_text3 = translate(text)
    arabic_text3 = arabic_text3[0]['translation_text']
    htranslated_text = f'<div dir="rtl">{arabic_text3}</div>'

    # Return the extracted text
    return text,htranslated_text

# Define the Gradio interface
text_recognition = gr.Interface(
    fn=extract_text,                    # The function that processes the image
    inputs=gr.Image(type="pil"),        # Input is an image (PIL format)
    outputs=[gr.Textbox(label='Extracted text'), gr.HTML(label= 'Translateted of Extracted text ')],   # Output is text
    title="Text Extraction and Translation | إستخراج النص وترجمتة",
    description="Upload an image then Submet to extract text and translate it to Arabic| قم برفع الصورة وأرسلها ليظهر لك النص من الصورة",
    examples =[["tx_image_0.png"]],
)

# Load trocr model for handwritten text extraction
processor = TrOCRProcessor.from_pretrained('microsoft/trocr-base-handwritten')
model = VisionEncoderDecoderModel.from_pretrained('microsoft/trocr-base-handwritten')

# Load MarianMT model for translation (English to Arabic)
translate = pipeline("translation",model="marefa-nlp/marefa-mt-en-ar")

def recognize_handwritten_text(image2):
  # process and and extract text
  pixel_values = processor(images=image2, return_tensors="pt").pixel_values
  generated_ids = model.generate(pixel_values)
  generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

  # Translate extracted text from English to Arabic
  arabic_text2 = translate(generated_text)
  arabic_text2 = arabic_text2[0]['translation_text']
  htranslated_text = f'<div dir="rtl">{arabic_text2}</div>'

  # Return the extracted text and translated text
  return generated_text, htranslated_text

# Gradio interface with image upload input and text output
handwritten_rec = gr.Interface(
    fn=recognize_handwritten_text,
    inputs=gr.Image(label="Upload Image"),
    outputs=[gr.Textbox(label='English Text'),
             gr.HTML(label='Arabic Text')],
    title="Handwritten Text Extraction | | إستخراج النص المكتوب بخط اليد وترجمتة",
    description="Upload an image then Submet to extract text and translate it to Arabic| قم برفع الصورة وأرسلها ليظهر لك النص من الصورة",
    examples =[["tx_image_1.png"]]
)

# Combine all interfaces into a tabbed interface
demo = gr.TabbedInterface([img_cap_en_ar, text_recognition, handwritten_rec], ["Extract_Caption", " Extract_Digital_text", " Extract_HandWritten_text"])
demo.launch(debug=True)