newensemble.py

import os
import numpy as np
import cv2
import matplotlib.pyplot as plt
from skimage.feature import hog
import joblib

# HOG Parameters
IMG_SIZE = (150, 150)
orientations = 9
pixels_per_cell = (8, 8)
cells_per_block = (2, 2)
block_norm = 'L2-Hys'
THRESHOLD = 0.5  # if ensemble genuine probability is >= 0.5, classify as "Not Fraud" (genuine)

# Function to extract HOG features from an image
def extract_hog_features(image_path):

    img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
    if img is None:
        raise ValueError(f"Error: Unable to load image at {image_path}")
    img = cv2.resize(img, IMG_SIZE)
    features = hog(img,
                   orientations=orientations,
                   pixels_per_cell=pixels_per_cell,
                   cells_per_block=cells_per_block,
                   block_norm=block_norm)
    return features.reshape(1, -1)

# Function to classify a signature using an ensemble of models
def ensemble_classify_signature(image_path, models, threshold=THRESHOLD):
    features = extract_hog_features(image_path)
    probs = []
    for model in models:
        prob = model.predict_proba(features)[0][1]  # probability for genuine class
        probs.append(prob)
    ensemble_prob = np.mean(probs)
    classification = "Not Fraud (Genuine)" if ensemble_prob >= threshold else "Fraud (Forged)"
    return classification, ensemble_prob, probs

def main():
    # Load the pre-trained models (update the paths as needed)
    knn_model_path = r"C:\Users\Vyse\Documents\GitHub\INF2008ML\knn_model.pkl"
    rf_model_path = r"C:\Users\Vyse\Documents\GitHub\INF2008ML\random_forest_model.pkl"
    svm_model_path = r"C:\Users\Vyse\Documents\GitHub\INF2008ML\svm_model.pkl"
    logreg_model_path = r"C:\Users\Vyse\Documents\GitHub\INF2008ML\writer_independent_logreg_model.pkl"
    adaboost_model_path = r"C:\Users\Vyse\Documents\GitHub\INF2008ML\adaboost_model.pkl"
    
    knn = joblib.load(knn_model_path)
    rf = joblib.load(rf_model_path)
    svm = joblib.load(svm_model_path)
    logreg = joblib.load(logreg_model_path)
    adaboost = joblib.load(adaboost_model_path)
    
    models = [knn, rf, svm, logreg, adaboost]
    
    # Update the test image path to the signature you wish to classify
    test_image_path = r"C:\Users\Vyse\Documents\GitHub\INF2008ML\signatures_cedar\unseen_data_for_testing\unseen_org\original_50_12.png"
    
    classification, ensemble_prob, individual_probs = ensemble_classify_signature(test_image_path, models, threshold=THRESHOLD)
    print("Ensemble Classification:", classification)
    print("Ensemble Genuine Probability: {:.2f}".format(ensemble_prob))
    print("Individual Model Probabilities:")
    print("KNN: {:.2f}".format(individual_probs[0]))
    print("Random Forest: {:.2f}".format(individual_probs[1]))
    print("SVM: {:.2f}".format(individual_probs[2]))
    print("Logistic Regression: {:.2f}".format(individual_probs[3]))
    print("Adaboost: {:.2f}".format(individual_probs[4]))
    
    # Optional: display the test image with its classification result
    img = cv2.imread(test_image_path, cv2.IMREAD_GRAYSCALE)
    if img is not None:
        img = cv2.resize(img, IMG_SIZE)
        plt.imshow(img, cmap="gray")
        plt.title(f"Test Image - Classified as: {classification}")
        plt.axis("off")
        plt.show()

if __name__ == "__main__":
    main()