|
Face Detection in a Pose Different than the One Learned AbstractDetecting upright faces is now a common task. On the other hand, detecting non upright faces is still problematic. By now, detecting faces in a given pose require a huge number of training images of faces in the same pose. To avoid this tedious and time consuming collect, we propose a solution based on an upright face detector that we modify to allow it to detect faces in another pose. To do so, we modify all subwindows positions which are used at detection time. Results show a clear performance improvement.
[Edit] |