Face Classification Based On Shape Features

Face recognition has been one of the most active areas in pattern recognition and computer vision in the past decades. This thesis focuses on the research on face classification. Automatic classifying the faces into different classes can accelerate the speed of recognition and also help to improve the accuracy of face recognition.By now, there are few literatures that had mentioned about the methods for automatic classifying face images by the shape features. This thesis proposes an approach which uses the ISODATA dynamic clustering algorithm based on Hausdorff distance and fuzzy C-mean method to classify the face shape features into different classes automatically.The thesis is mainly composed of two parts. First we locate the vital feature points and extract the shape features on human faces. Then, we present an approach to cluster the faces into different classes fast and reasonably,according to the shape features, such as face contour.A method based on human visual characteristics is described to find the 9 vital feature points on faces automatically, applying the geometry and symmetry of faces. This method can extract the features with properties of scale, translation and rotation invariance and locate feature points on faces exactly and quickly. By using SUSAN operator to locate the corner points on face and combining with directional integral projection, we provide a feasible way of locating the feature points on face images exactly. Additionally, we apply this result on the pretreatment of face images. A method to normalize face images using the distance of two far corners of eyes is presented. It could get more exact and steady results than those commonly using the distance of two irises.An automatic clustering method based on Hausdorff distance is proposed to classify the shape features on human faces, such as face contour, the shape of eyebrows, eyes, nose and mouth. We first extract all the feature points on face using a developed front-view based ASM algorithm. Then a classifying method derived from the ISODATA dynamic clustering algorithm is presented. It is more suitable to classify the shape features. Then the fuzzy C-mean method is introduced to make the classification results of faces more reasonable.The experiments show that the approaches presented in this thesis could automatic classify the face shape features fast and reasonably. Furthermore, we apply the methods to classify any kind of face shape features. Finally, the face classification results are applied to accelerate the speed of face recognition and improve the accuracy of face recognition.