This paper presents a real-time method for estimating the position and the size of a face in cluttered background. This method is based on the algorithm Condensation(Conditional Density Propagation) which was developed for tracking by stochastic propagation of conditional density. The proposed algorithm utilizes color cue as the measure of Condensation for the reduction of processing time. But it generates a problem of mismatch between color model and real face color owing to the variations of face color and the change of illumination. So I propose a new face color model. This model adapts to different face color and abrupt change of illumination by continuous update according to the current face color. Also a fast and simple method of segmenting face region was suggested. The proposed method operated at the speed of 13.6frame/sec in PC environment and has demonstrated highly reliable and accurate performance.
