A Robust Framework for Detection of Human Faces in Clutter Images

Automatic finding of human face from a complex background image is a challenging task in computer vision. This task is the first step in a large number of applications such as security control, human computer interface, face recognition, and image database management. This paper proposes a complete framework for face detection in clutter color images in presence of varying lighting conditions as well as presence of complex background. The proposed framework relies in using a group of neural network classifiers. These classifiers are applied sequentially to detect faces in the image. Initial skin regions over the entire image are detected using one of these classifiers by labeling the image pixels as skin/ non-skin pixels. Skin regions are classified into face/non-face candidates using the correlation between skin region and a prototype face-outline shape image. The facial features mouth and eye are detected within that face candidate using another two classifiers, one for each feature. Finally, the face candidate is verified by applying some face geometrical constrains. Experimental results demonstrate that the proposed framework can be used as a face detection tool over wide range of facial variations in color, position, scale, orientation and expressions.