Evaluation of 3D Facial Feature Selection for Individual Facial Model Identification

Face recognition using 3D information has been intensively investigated in recent years. The features selected from 3D facial surfaces are invariant to pose and lighting conditions. However, they are sensitive to expression variations. In this paper, we investigate the issues on selecting good features for 3D facial shape classification, and evaluate its applicability to various types of models. Based on our existing work on feature selection using a genetic algorithm, we derived a set of features from the individualized wire-frame models. We evaluate the usefulness of such features not only to the generated models from images, but also to the range data from 3D imaging systems with variable resolutions. We tested the algorithm on two types of data sets: generic model based dataset and range-scan model based dataset. Experimental results show that the optimal features derived from both datasets are robust among two databases. The resolution of captured models affects the selection of optimal features; however, the combination of the optimal features improves the recognition rate