Face registration and feature localization on depth data

Proper pre-processing of input data is a critical requirement in real time face identification and authentication systems that can improve its performance. Necessary steps involved depends on the nature input data. This paper focuses on pre-processing steps such as face registration and facial feature localization for facial depth data. Face registration algorithm is based on conventional ICP algorithm. Doing ICP with selected facial points can improve its efficiency. So corner points from 3D faces are considered for ICP to obtain registration parameters. Eye corner localization is based on the curvature analysis and nose tip detection is based on the principle that it is having highest depth value among all other facial components. Experiments are done with two different databases, low quality depth data from RGB-D FACE database and better quality FRAV3D.