Face detecting and tracking is the first step of the face informationprocessing. In this paper, studies on automatic face tracking in video imagesequences are carried out. A systematic summary of previous work is givenfirst. Then this paper presents a hierarchical face detection algorithm invideo image sequences under complex backgrounds and varying lightingconditions, which makes a good use of human face pattern's valuableinformation. The difficult detection task is divided into four steps: thepreprocessing, which is to gain motion regions with the adaptive threshold;the roughly detection, which is to gain candidate regions with humanskin-color model and morphological dilation-erosion operation; the finedetection with eyes and mouth color features' information; face locationwith eyes and mouth map projection. Experimental results demonstrate thatthe detection algorithm is efficient in complex background. Finally, thispaper also presents a new method for face tracking, which combinestemplate matching technique and verification of the matching resulttechnique based on the adaptive skin-color model. The tracking methoddeals with center moment matching of skin-color, motion prediction to theposition of the search region, verification of the matching result usingMahalanobis confidence measurement, and modeling the adaptiveskin-color. This face tracking system enables the robust tracking of facewith translation, rotation, zoom in/out, and has a special attention toocclusion handling. |