This paper presents a novel method for automatically classifying facial images based on labeled elastic graph matching, 2D Gabor wavelet representation, and linear discriminant analysis (LDA). The method is tested on three classification tasks: sex, "race," and facial expression. The algorithm combines the advantages of Gabor wavelet representation with the ability to train the system quickly from examples, similar to the Fisherface algorithm. The algorithm first registers a grid with the face and then classifies the face based on features extracted at grid points. The features are derived from a 2D Gabor wavelet transform, which provides a compromise between spatial and spatial frequency domain accuracy and is robust to small changes in grid node positions. The algorithm is tested on three image sets: one for sex, "race," and expression classification, and two for facial expression recognition. The results show that the algorithm achieves high classification rates, with generalization rates of 91% for expression, 95% for "race," and 92% for sex recognition. The algorithm is also tested on a set of facial expressions posed by nine Japanese females, achieving a generalization rate of 92%. The algorithm is found to be robust to shifts in node position and maintains generalization rates exceeding 90% for the three classification tasks. The algorithm is also tested on a set of facial expressions from Ekman and Friesen, achieving a peak generalization rate of 82%. The algorithm is found to be insensitive to color and not all expressions are recognized equally well. The saliency maps show that the regions around the eyes and mouth are most important for facial expression recognition. The algorithm is limited in that it can only extract categorical information about faces and is insensitive to color. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 75% for a novel expresser. The algorithm is compared to other methods and is found to be competitive. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 82% for the Ekman pictures. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 82% for the Ekman pictures. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 82% for the Ekman pictures.This paper presents a novel method for automatically classifying facial images based on labeled elastic graph matching, 2D Gabor wavelet representation, and linear discriminant analysis (LDA). The method is tested on three classification tasks: sex, "race," and facial expression. The algorithm combines the advantages of Gabor wavelet representation with the ability to train the system quickly from examples, similar to the Fisherface algorithm. The algorithm first registers a grid with the face and then classifies the face based on features extracted at grid points. The features are derived from a 2D Gabor wavelet transform, which provides a compromise between spatial and spatial frequency domain accuracy and is robust to small changes in grid node positions. The algorithm is tested on three image sets: one for sex, "race," and expression classification, and two for facial expression recognition. The results show that the algorithm achieves high classification rates, with generalization rates of 91% for expression, 95% for "race," and 92% for sex recognition. The algorithm is also tested on a set of facial expressions posed by nine Japanese females, achieving a generalization rate of 92%. The algorithm is found to be robust to shifts in node position and maintains generalization rates exceeding 90% for the three classification tasks. The algorithm is also tested on a set of facial expressions from Ekman and Friesen, achieving a peak generalization rate of 82%. The algorithm is found to be insensitive to color and not all expressions are recognized equally well. The saliency maps show that the regions around the eyes and mouth are most important for facial expression recognition. The algorithm is limited in that it can only extract categorical information about faces and is insensitive to color. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 75% for a novel expresser. The algorithm is compared to other methods and is found to be competitive. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 82% for the Ekman pictures. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 82% for the Ekman pictures. The algorithm is also found to be effective for facial expression recognition, with a generalization rate of 82% for the Ekman pictures.