Nose detection based feature extraction on both normal and abnormal 3D faces

This paper presents a feature extraction method, which does not require a frontal model and is applicable on both normal and some abnormal faces (abnormality caused by some genetic syndromes). The algorithm starts by a nose detection step. A geometry property match is employed to get possible candidates and then a symmetry calculation is carried out to select out the nose tip. Once the nose tip is decided, the 3D model is adjusted and normalized. After that, region segmentation is performed efficiently benefiting from the nose tip hint acquired in previous step. With the feature regions classified, feature points can be extracted easily with a set of profiles. Experiment results show that the overall performance is over 90% for both normal and abnormal models.