A quantitative comparison of 3D face databases for 3D face recognition

Abstract

During the last decade research in face recognition has shifted from 2D to 3D face representations. The need for 3D face data has resulted in the advent of 3D databases. In this paper, we first give an overview of publicly available 3D face databases containing expression variations, since these variations are an important challenge in today's research. The existence of many databases demands a quantitative comparison of these databases in order to compare more objectively the performances of the various methods available in literature. The ICP algorithm is used as baseline algorithm for this quantitative comparison for the identification and verification scenario, allowing to order the databases according to their inherent difficulty. Performance analysis using the rank 1 recognition rate for identification and the equal error rate for verification reveals that the FRGC v2 database can be considered as the most challenging. Therefore, we recommend to use this database further as reference database to evaluate (expression-invariant) 3D face recognition algorithms. As second contribution, the main factors that influence the performance of the baseline technique are determined and attempted to be quantified. It appears that (1) pose variations away from frontality degrade performance, (2) expression types affect results, (3) more intense expressions degrade recognition, (4) an increasing number of expressions decreases performance and (5) the number of gallery subjects degrades performace. A new 3D face recognition algorithm should be evaluated for all these factors.