The paper presents a method for automatically classifying facial images based on labeled elastic graph matching, 2D Gabor wavelet representation, and linear discriminant analysis (LDA). The method combines the advantages of Gabor wavelet representation and LDA, allowing for robust and quick training from examples. The algorithm is divided into two main steps: registration of a grid with the face and classification based on feature values extracted at grid points. The 2D Gabor wavelet representation captures spatial frequency structure while preserving spatial relations, making it robust to small changes in grid node positions. LDA is used to separate vectors into clusters with different facial attributes, such as sex, "race," and expression. Experiments using three image sets (sex, "race," and expression) demonstrate the system's robustness and generalization rates exceeding 90%. The paper also discusses the saliency of features, showing that the eyes and mouth are critical for determining facial expressions. The method is adaptable and can be trained quickly, making it suitable for real-time applications. However, it is limited to categorical information and does not handle color information.The paper presents a method for automatically classifying facial images based on labeled elastic graph matching, 2D Gabor wavelet representation, and linear discriminant analysis (LDA). The method combines the advantages of Gabor wavelet representation and LDA, allowing for robust and quick training from examples. The algorithm is divided into two main steps: registration of a grid with the face and classification based on feature values extracted at grid points. The 2D Gabor wavelet representation captures spatial frequency structure while preserving spatial relations, making it robust to small changes in grid node positions. LDA is used to separate vectors into clusters with different facial attributes, such as sex, "race," and expression. Experiments using three image sets (sex, "race," and expression) demonstrate the system's robustness and generalization rates exceeding 90%. The paper also discusses the saliency of features, showing that the eyes and mouth are critical for determining facial expressions. The method is adaptable and can be trained quickly, making it suitable for real-time applications. However, it is limited to categorical information and does not handle color information.