Automated Scene-Specific Selection of Feature Detectors for 3D Face Reconstruction
In comparison with 2D face images, 3D face models have the advantage of being illumination and pose invariant, which provides improved capability of handling changing environments in practical surveillance. Feature detection, as the initial process of rec
- PDF / 1,154,520 Bytes
- 12 Pages / 430 x 660 pts Page_size
- 20 Downloads / 193 Views
Abstract. In comparison with 2D face images, 3D face models have the advantage of being illumination and pose invariant, which provides improved capability of handling changing environments in practical surveillance. Feature detection, as the initial process of reconstructing 3D face models from 2D uncalibrated image sequences, plays an important role and directly affects the accuracy and robustness of the resulting reconstruction. In this paper, we propose an automated scene-specific selection algorithm that adaptively chooses an optimal feature detector according to the input image sequence for the purpose of 3D face reconstruction. We compare the performance of various feature detectors in terms of accuracy and robustness of the sparse and dense reconstructions. Our experimental results demonstrate the effectiveness of the proposed selection method from the observation that the chosen feature detector produces 3D reconstructed face models with superior accuracy and robustness to image noise.
1 Introduction The 3D reconstruction from uncalibrated video sequences has attracted increasing attention recently. Most of the proposed algorithms regarding feature matching and projective/metric reconstruction have applications in 3D reconstruction of man-made scenes [1, 2]. Recently, because of the difficulties in 2D face recognition caused by illumination and pose variations, recognition algorithms using 3D face models have emerged [3], which calls for reconstruction algorithms designed particularly for faces. Hu et al. used salient facial feature points to project a 2D frontal view image onto a 3D face model automatically [4] and illustrated improved face recognition rates using the 3D model despite pose and illumination variations. Chowdhury et al. reconstructed 3D facial feature points and obtained a 3D face model by fitting these points to a generic 3D face model [5]. Most existing 3D reconstruction algorithms start with feature selection and matching [1, 2, 5]. Based on the matched features in consecutive frames, 3D projective and metric structures are recovered. Therefore, the accuracy and robustness of feature detection and matching directly affect the overall performance of the reconstruction. Popular features for 3D reconstruction are image corners and lines. For a man-made scene, there exist well-defined corners, which facilitate the use of fast and G. Bebis et al. (Eds.): ISVC 2007, Part I, LNCS 4841, pp. 476–487, 2007. © Springer-Verlag Berlin Heidelberg 2007
Automated Scene-Specific Selection of Feature Detectors for 3D Face Reconstruction
477
straightforward feature detectors such as Harris corners. However, for face images, corners are not as distinguishable as in man-made scenes. In addition, face images include smooth areas, for example cheek and forehead, where feature matching becomes more ambiguous. Therefore, it is important to find an appropriate feature detector, which can make full use of facial features and avoid smooth areas simultaneously for 3D face reconstruction. In this paper, we propo
Data Loading...
