import gradio as gr
import pandas as pd
import numpy as np
import joblib
import os
from skimage.measure import shannon_entropy
from skimage.color import rgb2hsv
from scipy.ndimage import generic_filter
import cv2

# Function to extract features from the image
def extract_features(image_path):
    # Load image and convert to RGB
    image = cv2.imread(image_path)
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    
    # Extract RGB means
    meanr = np.mean(image[:, :, 0])  # Red channel
    meang = np.mean(image[:, :, 1])  # Green channel
    meanb = np.mean(image[:, :, 2])  # Blue channel
    
    # Convert to HSI and compute HHR
    hsv_image = rgb2hsv(image)
    hue = hsv_image[:, :, 0]
    high_hue_pixels = np.sum(hue > 0.95)
    total_pixels = hue.size
    HHR = high_hue_pixels / total_pixels
    
    # Convert to Grayscale
    gray_image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
    
    # Compute Entropy
    Ent = shannon_entropy(gray_image)
    
    # Compute Brightness
    B = np.mean(gray_image)
    
    # Sliding window for gray-level features
    def g1_filter(window):
        return window[4] - np.min(window)

    def g2_filter(window):
        return np.max(window) - window[4]

    def g3_filter(window):
        return window[4] - np.mean(window)

    def g4_filter(window):
        return np.std(window)

    def g5_filter(window):
        return window[4]

    # Apply filters with 3x3 window
    g1 = generic_filter(gray_image, g1_filter, size=3).mean()
    g2 = generic_filter(gray_image, g2_filter, size=3).mean()
    g3 = generic_filter(gray_image, g3_filter, size=3).mean()
    g4 = generic_filter(gray_image, g4_filter, size=3).mean()
    g5 = generic_filter(gray_image, g5_filter, size=3).mean()

    # Return features
    return {
        "meanr": meanr,
        "meang": meang,
        "meanb": meanb,
        "HHR": HHR,
        "Ent": Ent,
        "B": B,
        "g1": g1,
        "g2": g2,
        "g3": g3,
        "g4": g4,
        "g5": g5,
    }

# Function to check if the image is a valid file format
def check_image_format(image):
    try:
        img = cv2.imread(image)
        if img is not None:
            return True
        else:
            return False
    except Exception as e:
        return False

# Function to predict hemoglobin value
def predict_hemoglobin(age, gender, image):
    print(f"Image path: {image}")  # Debugging line to check the image path

    # Check if the image file is valid
    if not check_image_format(image):
        return "Error: The uploaded image file is not recognized or is corrupt."
    
    # Extract features from the image
    features = extract_features(image)
    
    # Ensure gender is encoded correctly (0 for female, 1 for male)
    features['Gender'] = 1 if gender.lower() == 'M' else 0
    features['Age'] = age
    
    # Create a DataFrame for features (do not include Hgb, as it's the predicted value)
    features_df = pd.DataFrame([features])

    # Load the trained model, scaler, and label encoder
    svr_model = joblib.load('svr_model (1).pkl')
    scaler = joblib.load('minmax_scaler.pkl')
    label_encoder = joblib.load('label_encoder.pkl')

    # Ensure that features_df matches the expected training feature set (without 'Hgb')
    expected_columns = ['meanr', 'meang', 'meanb', 'HHR', 'Ent', 'B', 'g1', 'g2', 'g3', 'g4', 'g5', 'Age', 'Gender']
    
    for col in expected_columns:
        if col not in features_df:
            features_df[col] = 0  # Or some default value to match the expected columns.
    
    features_df = features_df[expected_columns]  # Ensure the correct order of columns
    
    # Apply scaling (do not include 'Hgb' as it is the target)
    features_df_scaled = scaler.transform(features_df)
    
    # Predict hemoglobin using the trained SVR model
    hemoglobin = svr_model.predict(features_df_scaled)[0]
    
    return f"Predicted Hemoglobin Value: {hemoglobin:.2f}"

# Gradio Interface setup
def create_gradio_interface():
    # Define the inputs and outputs for the Gradio interface
    image_input = gr.Image(type="filepath", label="Image (Path to Image)", interactive=True)
    age_input = gr.Number(label="Age", value=25, precision=0)
    gender_input = gr.Radio(choices=["Male", "Female"], label="Gender", value="Male")
    
    # Set up the Gradio interface with the prediction function
    gr.Interface(fn=predict_hemoglobin,
                 inputs=[age_input, gender_input, image_input],
                 outputs="text",
                 live=True).launch(share=True)

# Run the Gradio app
if __name__ == "__main__":
    create_gradio_interface()