In this paper, we present a novel method to detect a human skin region from a given head and shoulder image. The presented method consists of two stages: region segmentation and facial region detection. In the region segmentation, the input image is segmented into an appropriate set of arbitrary regions using the wavelet-based watershed algorithm. Then, to merge the regions forming an object, we use a spatial similarity between two regions since the regions forming an object share some common wavelet characteristics. In facial region detection, the facial regions are identified from the segmented results using a skin-color model. The results of region segmentation and facial region detection are integrated to provide facial regions with accurate and closed boundaries. In our experiments, the algorithm detected 87-94% of the faces, including frames from videoconference sequences. The average run time range from 0.23-0.34 sec per frame.