A face clustering method based on facial shape information

A face clustering method based on facial shapes is proposed in this paper. In order to improve image retrieval efficiency in large database, a modified ISODATA clustering method is used to categorize faces into different classes using facial shape features. Firstly, face images are clustered into 7 classes according to face contour; secondly, in each face contour class, facial shape features including the eyes, nose, mouth and the relative location of face organs are used to further cluster faces into several small classes. The facial shape features are extracted by a modified ASM method. The Hausdorff distance is applied to calculate distance of two facial shape point sets. Experimental results demonstrate that the proposed approach could classify human faces reasonably and accurately. Moreover, it could help to accelerate the speed of face recognition.