The image blending method for face swapping

Abstract

Recently, image synthesis technology such as face swapping have been receiving attention gradually. In this issue, the rigid boundary line which is occurred by the large difference of color and brightness between reference and target image in synthesized image is a serious problem. This boundary line is the main cause of unnatural face swapping. In this paper, we propose a blending method which can deal with this problem by adaptive weight value and require low-cost for computation. In experiment, we verify that proposed method can remove the boundary line effectively during the image synthesis and preserve the principal parts of human face such as eyes, nose and mouth. Furthermore, we demonstrate that our method can compute the algorithm with 600 by 600 resolution image in near realtime.