Face Detection on Still Images Using HIT Maps

Abstract

We present a fully automatic solution to human face detection on still color images and to the closely related problems of face segmentation and location. Our method is based on the use of color and texture for searching skin-like regions in the images. This is accomplished with connected component analysis in adaptatively thresholded images. Multiple candidate regions appear, so determining whether each one corresponds or not to a face, solves the detection problem and allows a straightforward segmentation. Then, the main facial features are located using accumulative projections. We present some results on a database of typical TV and videoconference images. Finally, we extract some conclusions and advance our future work.