The proposed database has been used for testing a face
detection algorithm similar to the one proposed in [3]. Here
the depth information is first used to find the closest object
to the camera, which is a valid assumption for most facial
analysis systems as it is usually the case that a user stands
in front of the camera in these systems. Having reduced the
search space of the input image, again depth information is
used to find some face candidate regions. The depth image
data may contain points or even areas with undefined depth.
These areas are usually small but they can also be fairly
large and can seriously affect the result of the face detection.
Therefore, it should be first checked if there is any such holes
in the user’s face. If so, they need to be filled up. Here a
mean filter of size 13×13 has been used for this purpose.
Figure 3(b) shows the results of applying such a filter to the
face image shown in Figure 3(a).