|
In face recognition, the majority of the research are use 2D image information to extract facial features and identification. In recent years, 3D image and point clouds have been used for identification, however the 3D modality is usually needed to be built before face recognition. In addition, the process of the 3D modality is time-consuming and complicated. 3D point clouds could be directly used for face recognition without 3D modality. The myopia is up to one million population in Taiwan, people sometimes would wear glasses while 3D scanning. In this study, influence of glasses-wearing on face recognition will be taken into account. The different kinds of glasses, such as reflected and non-reflected types, are considered. Impact of the glasses-wearing on the extraction of nose tip and root will be discussed. On the process of 3D point clouds, five feature values after the nose tip and root were found out are determine: (1) distance from nose root to nose tip, (2) distance of nose tip to vertical-section fitted line, (3) distance of nose root to vertical-section fitted line, (4) distance of nose tip to cross-section fitted line, and (5) distance of nose root to cross-section fitted line. With these five values, the influence of glasses-wearing on the face recognition will be discussed. In this study, three methods are introduced to preliminarily remove the 3D point clouds of glasses frame. The methods are: (1) manual deletion, (2) mean method, and (3) minimum method. Of these three methods, the best result is obtained from the manual deletion, but it is time-consuming. The mean and minimum method are rapid and efficient in comparison with the manual deletion. However, recognition success rate is not good enough. Therefore, it is suggested that the glasses have to be taken off before 3D scanning in order to obtain good recognition success rate under the circumstances that the removal of 3D point clouds of glasses frame does not work properly.
|