VIDEO SIGNAL FACE REGION DETECTION

摘要

A highly efficient method is disclosed for rapidly detecting human face regions in color images and video frames. The method consits of three stages, where chrominance (stage 1), shape (stage 2) and spatial frequency information (stage 3) are used respectively. In the first stage, the chrominance data for incremental image portions is compared to skin tone data to identify skin tone image portions. In the second stage regions of contiguous skin tone image portions are compared to templates corresponding to the shape of faces and images. In the optional third stage, the spatial frequency characteristics of luminance data in the selected possible face region is compared to at least one threshold value to eliminate regions that do not include spatial frequency characteristics corresponding to those of a face region.