Abstract:

Traditionally, AdaBoost based face detection algorithms use frontal faces to train classifiers, which cannot detect multipose faces with arbitrary rotations. Noticing that the eye structures vary less than face structures in multipose faces, eyes were used to train the classifier in AdaBoost algorithm, and present a fast algorithm to detect multipose faces in complex scenes. An image is first mapped to a YCrCb color space and the skin region is detected with the skin color distribution. The eye region is then detected from the skin region via the AdaBoost algorithm. The face is located according to the eye positions relative to faces. Experimental results show that this algorithm performs well for the detection of multipose faces.

Key words: color space, skin model, multi-pose face detection, eye detection