Evaluating the Performance of 3D Face Reconstruction Algorithms

The use of 3D data in face image processing applications received increased attention during the recent years. However, the development of efficient 3D face processing applications (i.e face recognition) relies on the availability of appropriate 3D scanning technology that enables real time, accurate, non-invasive and low cost 3D data acquisition. 3D scanners currently available do not fulfill the necessary criteria. As an alternative to using 3D scanners, 3D face reconstruction techniques, capable of generating a 3D face from a single or multiple face images, can be used. Although numerous 3D reconstruction techniques were reported in the literature so far the performance of such algorithms was not evaluated in a systematic approach. In this paper we describe a 3D face reconstruction performance evaluation framework that can be used for assessing the performance of 3D face reconstruction techniques. This approach is based on the projection of a set of existing 3D faces into 2D using different orientation parameters, and the subsequent reconstruction of those faces. The comparison of the actual and reconstructed faces enables the definition of the reconstruction accuracy and the investigation of the sensitivity of an algorithm to different conditions. The use of the proposed framework is demonstrated in evaluating the performance of two 3D reconstruction techniques.