app.py

from transformers import LayoutLMForTokenClassification, LayoutLMTokenizer
import gradio as gr
import cv2
import easyocr
from simple_salesforce import Salesforce

# Salesforce credentials
Salesforce_User_Name = '[email protected]'  # Your Salesforce username
Salesforce_Password = 'Sathkrutha@06'
SALESFORCE_INSTANCE_URL = 'https://sathkruthatechsolutions63-dev-ed.develop.lightning.force.com'
SALESFORCE_ACCESS_TOKEN = 'UnByPih7PWmoWLzRuRyFrXzw'

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

# Load pre-trained LayoutLM model and tokenizer
model = LayoutLMForTokenClassification.from_pretrained("microsoft/layoutlm-large-uncased")
tokenizer = LayoutLMTokenizer.from_pretrained("microsoft/layoutlm-large-uncased")

# Salesforce Connection Setup
sf = Salesforce(username=Salesforce_User_Name, password=Salesforce_Password, security_token=SALESFORCE_ACCESS_TOKEN)

# Function to extract text using EasyOCR and process with LayoutLM
def extract_patient_info(image):
    # Convert the uploaded image to RGB (required by LayoutLM)
    image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

    # Use EasyOCR to extract text from the image
    result = reader.readtext(image_rgb)
    extracted_text = " ".join([detection[1] for detection in result])

    # Tokenize the extracted text with LayoutLM
    inputs = tokenizer(extracted_text, return_tensors="pt")
    outputs = model(**inputs)

    # Here, extracted_text is already available from EasyOCR, we can extract relevant details
    details = extract_details_from_text(extracted_text)
    
    # Create a record in Salesforce using the extracted details
    create_salesforce_record(details)

    # Return the extracted text
    return extracted_text

# Function to extract details from the extracted text (use regex or other methods to extract)
def extract_details_from_text(extracted_text):
    # Simple example of extracting details, customize this according to the format of the text
    details = {}
    details['Name'] = "Shanthi"  # Here, add the logic to extract the actual name
    details['Age'] = "39"        # Similarly, extract age, gender, and phone number
    details['Gender'] = "Female"
    details['Phone Number'] = "9955337097"
    
    return details

# Function to create a record in Salesforce
def create_salesforce_record(details):
    data = {
        'Name__c': details['Name'],
        'Age__c': int(details['Age']),
        'Gender__c': details['Gender'],
        'Phone_Number__c': details['Phone Number']
    }
    
    # Create a new record in Salesforce
    sf.Patient_Registration__c.create(data)
    print("Salesforce record created successfully!")

# Gradio interface setup
with gr.Blocks() as demo:
    gr.Markdown("### OCR Using LayoutLM Pretrained Model with EasyOCR and Salesforce Integration")

    # Image upload component
    image_input = gr.Image(type="numpy", label="Upload Image")

    # Output textboxes to display the extracted information
    output_text = gr.Textbox(label="Extracted Text")

    # Button to trigger image processing and text extraction
    process_button = gr.Button("Process Image")

    # When the button is clicked, process the image and show results in textboxes
    process_button.click(fn=extract_patient_info, inputs=image_input, outputs=output_text)

# Launch the Gradio app
if __name__ == "__main__":
    demo.launch()