The main problem of face recognition technology is its relatively low recognition rate. In an attempt to improve the situation, NIST set up the Face Recognition Grand Challenge (FRGC)[1] in 2004 which aims to improve performance of face recognition systems by an order of magnitude. NIST has created large datasets of 2D images, 3D facial meshes and multiple 2D images.

[1] P. J. Phillips, P. Flynn, T.Scruggs, K.W. Bowyer, J. Chang, K. Hoffman, J. Marques, J. Min, and W. Worek. Overview of the Face Recognition Grand Challenge. In Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR05), San Diego, CA, June 2005.

For the purposes of our research, we haveR compiled a multimodal database containing visible images, IR images,and 3D data. The data were acquired during 2004 and containing a total of 356 datasets.

The UHDB1 database consists of 3D captures of 141 subjects. The 3D captures consist of 3D meshes and textures taken of the subject in 5 different poses (-90, -45, 0, 45, 90 degree yaw). The same subject is also captured in two different cars with the camera 90 degrees to the left of the subject. In normal pose, the subject appears to look to the left side of the image. Only half of the subject is visible in normal pose. In each car, each subject has 7 captures of different pose in a neutral expression and 1 capture of normal pose with a happy expression. In summary, a total of 21 captures are made per subject.

In order to analyze the impact of the variation in both pose and lighting, we acquired data from 23 subjects under 6 illumination conditions. For each illumination condition, we asked the subject to face four different points inside the room. This generated rotations on the Y axis. For each rotation on Y, we also acquired 3 images with rotations on the Z axis (assuming that the Z axis goes from the back of the head to the nose, and that the Y axis is the vertical axis through the subject’s head). Thus, we acquired images under 6 illumination conditions, 4 Y rotations and 3 Z rotations per subject. In total, we have 72 pose/light variation per subject for 23 subject. Each capture consists of both 2D image captured using a Canon (TM) DSLR camera and a 3d mesh captured by 3dMD (TM) 2-pod optical 3D system.

The 3D data were captured by a 3dMD(TM) two-pod optical scanner, while the 2D data were captured by a commercial Canon(TM) DSLR camera. The system has six diffuse lights that allow the variation of the lighting conditions. For each subject there is a single 3D scan (and the associated 2D texture) that is used as a gallery dataset and several 2D images that are used as probe datasets. All 2D images have one of the six possible lighting conditions. There are 26 subjects, resulting in 26 gallery datasets (3D plus 2D) and 800 probe datasets (2D only)

The UHDB2 database consists of 201 captures of subject ears with our 3DMD camera.

Request a Database