Region Based 3D Face Recognition

Abstract

This paper focuses on a region based methodology for expression in sensitive 3D face recognition process. Considering facial regions that are comparatively unchanging during expressions, results shows that using fifteen sub regions on the face can attain high 3D face recognition. We use a modified face recognition algorithm along with hierarchical contour based image registration for finding the similarity score. Our method operates in two modes: verification mode and confirmation mode. Crop 100 mm of frontal face region, apply preprocessing and automatically detect nose tip, translate the face image to origin and crop fifteen sub regions. The cropped sub regions are defined by cuboids which occupy more volumetric data, Nose Tip is the most projecting point of the face with the highest value along Z-axis so consider it as origin. The modified face recognition algorithm reduces the effects caused by facial expressions and artifacts. Finally a Hierarchical contour based image registration technique is applied which yields better results. The approach is applied on Bosphorus 3D datasets and achieved a verification rate of 95.3% at 0.1% false acceptance rate. In the identification scenario 99.3% rank one recognition is achieved.