texture_classification.py

from PIL import Image
import numpy as np
from skimage.feature import local_binary_pattern, graycomatrix, graycoprops
from sklearn.svm import LinearSVC
import os
from sklearn.metrics import accuracy_score, precision_score, \
    classification_report, confusion_matrix
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split
import joblib

IMAGE_SIZE_GLCM = 256
IMAGE_SIZE_LBP = 128

# LBP parameters
RADIUS = 1
N_POINTS = 8 * RADIUS
LBP_METHOD = "uniform"


def compute_glcm_histogram_pil(image, distances=[1], angles=[0], levels=8,
                               symmetric=True):

    # Convert the PIL image to a NumPy array
    image_np = np.array(image)

    # Quantize the grayscale image to the specified number of levels
    image_np = (image_np * (levels - 1) / 255).astype(np.uint8)

    # Compute the GLCM using skimage's graycomatrix function
    glcm = graycomatrix(image_np,
                        distances=distances,
                        angles=angles,
                        levels=levels,
                        symmetric=symmetric,
                        normed=True)

    # Extract GLCM properties
    homogeneity = graycoprops(glcm, 'homogeneity')[0, 0]
    correlation = graycoprops(glcm, 'correlation')[0, 0]

    # Create the feature vector
    feature_vector = np.array([homogeneity, correlation])

    return feature_vector


def image_resize(img, n):
    # Crop the image to a square by finding the minimum dimension
    min_dimension = min(img.size)
    left = (img.width - min_dimension) / 2
    top = (img.height - min_dimension) / 2
    right = (img.width + min_dimension) / 2
    bottom = (img.height + min_dimension) / 2
    img = img.crop((left, top, right, bottom))
    img = img.resize((n, n))
    return img


def get_lbp_hist(gray_image, n_points, radius, method):
    # Compute LBP for the image
    lbp = local_binary_pattern(gray_image, n_points, radius, method)

    # Compute LBP histogram
    lbp_hist, _ = np.histogram(lbp.ravel(), bins=np.arange(0, n_points + 3),
                               range=(0, n_points + 2))

    # Normalize the histogram
    lbp_hist = lbp_hist.astype("float")
    lbp_hist /= (lbp_hist.sum() + 1e-6)  # Normalized histogram
    return lbp_hist


def get_features(input_folder, class_label, method):
    data = []
    labels = []
    filenames = []
    image_files = [f for f in os.listdir(input_folder) if f.lower().endswith((
        '.png', '.jpg', '.jpeg', '.bmp', '.tiff'))]

    print(f"Total images found: {len(image_files)}")

    for _, file_name in enumerate(sorted(image_files)):
        img_path = os.path.join(input_folder, file_name)
        try:
            img = Image.open(img_path)
            img.verify()
            img = Image.open(img_path)
            img_gray = img.convert("L")

            if method == "GLCM":
                img_resized = image_resize(img_gray, IMAGE_SIZE_GLCM)
                hist = compute_glcm_histogram_pil(img_resized)
            elif method == "LBP":
                img_resized = image_resize(img_gray, IMAGE_SIZE_LBP)
                hist = get_lbp_hist(np.array(img_resized), N_POINTS, RADIUS,
                                    LBP_METHOD)

            data.append(hist)
            labels.append(class_label)
            filenames.append(file_name)  # Store the filenames

        except (FileNotFoundError, PermissionError) as file_err:
            print(f"File error with {file_name}: {file_err}")
        except Image.UnidentifiedImageError:
            print(f"Unidentified image file: {file_name}. Skipping this file.")
        except Exception as e:
            print(f"Unexpected error processing {file_name}: {e}")

    return data, labels, filenames


def main():

    # Set method
    method = "LBP"

    # Define paths
    grass_data, grass_labels, grass_filenames = get_features(
        "./raw_data/raw_grass_dataset", "Grass", method)
    wood_data, wood_labels, wood_filenames = get_features(
        "./raw_data/raw_wood_dataset", "Wood", method)
    data = grass_data + wood_data
    labels = grass_labels + wood_labels
    filenames = grass_filenames + wood_filenames  # Combine filenames

    # Train-test split
    X_train, X_test, y_train, y_test, train_filenames, test_filenames = \
        train_test_split(data, labels, filenames, test_size=0.3,
                         random_state=9, stratify=labels)

    # Train the model
    model = LinearSVC(C=100, loss="squared_hinge")
    model.fit(X_train, y_train)

    # Make predictions on the test set
    y_pred = model.predict(X_test)

    # Calculate accuracy and precision
    accuracy = accuracy_score(y_test, y_pred)
    precision = precision_score(y_test, y_pred, average='macro')

    # Print the results
    print(f"Accuracy: {accuracy:.2f}")
    print(f"Precision: {precision:.2f}")

    # Get a classification report for additional metrics
    print("\nClassification Report:")
    print(classification_report(y_test, y_pred))
    # print(f"Radius: {RADIUS}, N: {N_POINTS}")

    # Calculate the confusion matrix
    conf_matrix = confusion_matrix(y_test, y_pred)

    # Print the confusion matrix
    print("Confusion Matrix:")
    print(conf_matrix)

    # Create a heatmap for visualization
    plt.figure(figsize=(6, 4))
    sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues',
                xticklabels=["Grass", "Wood"], yticklabels=["Grass", "Wood"])
    plt.xlabel('Predicted')
    plt.ylabel('True')
    plt.title('Confusion Matrix')
    plt.show()

    # Identify misclassified images
    misclassified = [fname for i, fname in enumerate(test_filenames)
                     if y_test[i] != y_pred[i]]

    print("Misclassified Images:")
    for fname in misclassified:
        print(fname)

    # Save model parameters for deployment
    joblib.dump(model, method + '_model.joblib')


if __name__ == "__main__":
    main()