pidoko/textureClassification

1import logging2from typing import Tuple3import numpy as np4import pandas as pd5import cv26import matplotlib.pyplot as plt7import seaborn as sns8from pathlib import Path9from sklearn.model_selection import train_test_split, GridSearchCV10from sklearn.preprocessing import StandardScaler11from sklearn.ensemble import RandomForestClassifier12from sklearn.svm import SVC13from sklearn.neighbors import KNeighborsClassifier14from sklearn.linear_model import LogisticRegression15from sklearn.metrics import accuracy_score, classification_report, confusion_matrix16import gradio as gr17from config import OUTPUT_DIR, IMAGE_SIZE, DISTANCES, ANGLES, LBP_RADIUS, LBP_POINTS, LBP_METHOD18 19# Setup Logging20logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")21 22# Paths to feature datasets23glcm_csv_path = Path(OUTPUT_DIR) / "texture_features_glcm.csv"24lbp_csv_path = Path(OUTPUT_DIR) / "texture_features_lbp.csv"25 26 27def load_datasets() -> Tuple[pd.DataFrame, pd.DataFrame]:28    """Load GLCM and LBP datasets separately with validation."""29    if not glcm_csv_path.exists() or not lbp_csv_path.exists():30        raise FileNotFoundError("One or both dataset files (GLCM or LBP) are missing.")31 32    df_glcm = pd.read_csv(glcm_csv_path)33    df_lbp = pd.read_csv(lbp_csv_path)34 35    return df_glcm, df_lbp36 37 38def preprocess_data(df: pd.DataFrame):39    """Splits data into train/test sets and applies standardization."""40    X = df.drop(columns=["label"])41    y = df["label"]42 43    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42, stratify=y)44 45    scaler = StandardScaler()46    X_train_scaled = scaler.fit_transform(X_train)47    X_test_scaled = scaler.transform(X_test)48 49    return X_train_scaled, X_test_scaled, y_train, y_test, scaler50 51 52def train_models(X_train, y_train):53    """Trains multiple classifiers using GridSearchCV."""54    param_grids = {55        "Random Forest": {56            "model": RandomForestClassifier(random_state=42),57            "params": {"n_estimators": [50, 100, 200], "max_depth": [None, 10, 20]},58        },59        "SVM": {60            "model": SVC(random_state=42),61            "params": {"C": [0.1, 1, 10], "kernel": ["linear", "rbf"]},62        },63        "k-NN": {64            "model": KNeighborsClassifier(),65            "params": {"n_neighbors": [3, 5, 7], "weights": ["uniform", "distance"]},66        },67        "Logistic Regression": {68            "model": LogisticRegression(max_iter=1000, random_state=42),69            "params": {"C": [0.1, 1, 10]},70        },71    }72 73    best_models = {}74    confusion_matrices = {}75 76    for name, config in param_grids.items():77        logging.info(f"Training {name}...")78        grid_search = GridSearchCV(config["model"], config["params"], cv=5, scoring="accuracy", n_jobs=-1)79        grid_search.fit(X_train, y_train)80 81        best_models[name] = grid_search.best_estimator_82 83    return best_models84 85 86def plot_confusion_matrices(y_test, models, X_test, filename_prefix):87    """Plots and saves confusion matrices for multiple models."""88    for name, model in models.items():89        y_pred = model.predict(X_test)90        cm = confusion_matrix(y_test, y_pred)91 92        plt.figure(figsize=(6, 5))93        sns.heatmap(cm, annot=True, fmt="d", cmap="Blues", xticklabels=np.unique(y_test), yticklabels=np.unique(y_test))94        plt.title(f"Confusion Matrix - {name}")95        plt.xlabel("Predicted")96        plt.ylabel("Actual")97 98        output_path = Path(OUTPUT_DIR) / f"{filename_prefix}_conf_matrix_{name}.png"99        plt.savefig(output_path)100        plt.close()101 102        logging.info(f"Confusion matrix for {name} saved to {output_path}")103 104 105def classify_texture(image, feature_type, model_name):106    """Classifies an input image using the selected feature type and model."""107    image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)108    image = cv2.equalizeHist(image)109    image = cv2.resize(image, IMAGE_SIZE)110 111    if feature_type == "GLCM":112        from feature import compute_glcm_features  # Import only when needed113 114        features = compute_glcm_features(image).reshape(1, -1)115        features = scaler_glcm.transform(features)116        prediction = best_models_glcm[model_name].predict(features)[0]117 118    elif feature_type == "LBP":119        from feature import compute_lbp_features120 121        features = compute_lbp_features(image).reshape(1, -1)122        features = scaler_lbp.transform(features)123        prediction = best_models_lbp[model_name].predict(features)[0]124 125    return prediction126 127try:128    # Load datasets129    df_glcm, df_lbp = load_datasets()130 131    # Preprocess GLCM and LBP datasets132    X_train_glcm, X_test_glcm, y_train_glcm, y_test_glcm, scaler_glcm = preprocess_data(df_glcm)133    X_train_lbp, X_test_lbp, y_train_lbp, y_test_lbp, scaler_lbp = preprocess_data(df_lbp)134 135    # Train models separately for GLCM and LBP136    best_models_glcm = train_models(X_train_glcm, y_train_glcm)137    best_models_lbp = train_models(X_train_lbp, y_train_lbp)138 139    # Plot confusion matrices140    plot_confusion_matrices(y_test_glcm, best_models_glcm, X_test_glcm, "GLCM")141    plot_confusion_matrices(y_test_lbp, best_models_lbp, X_test_lbp, "LBP")142 143    # Define Hugging Face App Title144    title = "Texture Classification Using GLCM and LBP"145 146    title += "\n\nUpload image of a texture, choose a feature extraction method, and pick a classifier to predict."147 148    title += "\n\nAppropriate image has little to no noise (only relevant texture)"149 150    # Define path to sample images151    sample_images = [152        ["sample_images/wood1.jpg", "GLCM", "SVM"],153        ["sample_images/brick1.jpg", "LBP", "Random Forest"],154        ["sample_images/stone1.jpg", "GLCM", "k-NN"]155    ]156    157    # Gradio Interface158    interface = gr.Interface(159        fn=classify_texture,160        inputs=[161            gr.Image(type="numpy"),162            gr.Radio(["GLCM", "LBP"], label="Feature Type"),163            gr.Dropdown(choices=list(best_models_glcm.keys()), label="Select Classifier"),164        ],165        outputs=gr.Label(),166        title=title,167        examples=sample_images168    )169 170    logging.info("Launching Gradio interface...")171    interface.launch()172 173except Exception as e:174    logging.error(f"An error occurred: {e}")175    exit(1)176