Facial Feature Extraction and Change Analysis Using Photometric Stereo

摘要

This paper presents a new technique for three-dimensional face analysis aimed towards improving the robustness of face recognition. All of the 3D data used in the paper are obtained from a high-speed photometric stereo arrangement. First, a nose detection algorithm is presented, which is largely based on existing work, before a novel method for finding the nasion is described. Both of these methods rely solely on the 3D data. A new eye detection method is then described that uses a combination of 3D and 2D information with adaptive thresholding applied to the region of the image surrounding the eyes. The next main contribution of the paper is an analysis of the effects of makeup and facial hair on the success of the reconstruction and feature detection. We found that our method is very robust to such complications and can also handle spectacles and pose variation in many cases.