Abstract
A video transcoding method of facial assistance information is disclosed, wherein a Gaussian distribution model is used first to simulate the probability density distribution of skin color for determining the probability threshold. Then each pixel in the frame is put into the Gaussian distribution model and the probability density distribution is calculated. Then the probability density function value is compared with the probability threshold, so as to form a binary image according to the comparison result. A double integration projection is performed to the binary image, so that the top border and the bottom border of facial block can be determined. A vertical integration projection is performed to the region of binary image between two borders, so that the left border and the right border of facial block can be determined. After distinguishing the facial region, a dynamic error weight adjusting mechanism is employed for the rate control of macro block level. The method is to give more bits for facial block, and the amount of data used is reduced for the non-facial region and little movement. Thus, with a little sacrificing, the quality of background picture and the quality of facial region can be greatly enhanced. Since the facial part of the members who join the meeting is the emphasized point during the video conference, the visual feeling can be greatly enhanced by enhancing the quality of facial region, and the sacrifice of background picture is not easily sensed.