Face Detection Based on Improved Skin Model and Local Iterated Conditional Modes

The use of skin model is a common way for face detection in color image. However, in the actual environments, the detection accuracy of skin model always declines as a result of illumination variations, complex backgrounds and high overlap between skin and non-skin regions. To reduce the false-positive and false-negative rate of face detection, a hybrid method of coarse-to-fine face detection is proposed. In the course of coarse localization, an improved YCbCr color space-based skin model is established to extract the human skin regions and the connected components analysis is used to define the face region. By this way, most non-skin backgrounds are eliminated and only face and skin-like regions are preserved. The fine face detection employs local iterated conditional modes within the candidate region to precisely localize the real face contour, and the ensuant morphological operations are used to the remove burrs and holes from the face region. Experiments show that this method performs well for complex color image and the accuracy rate of face detection is high with less false detection.