Research On Video Face Editing Algorithm Based On Possion Blending

In the field of computer vision and image processing,video face editing is widely used in social entertainment and teleconference.As a new media processing method,video face editing plays an important role in the film industry.It has become a research topic of universities and research institutes.This paper proposes an automatic face editing technology in video,including editing and processing the whole face and local facial area.The technology does not need complex 3D image scanning equipment or 3D model database for assistance.It completes the editing operation of the face on the basis of the two-dimensional image,and does not need the user to perform the human-computer interaction.This technique automatically handles the two clips of input video.The face editing method proposed in this paper mainly consists of five parts,including face detection and landmark location,video face registration,mouth shape matching,optimal mask generation and image fusion.This article mainly has the following three aspects of innovation.A fast video face registration method is proposed,which is mainly used to eliminate the jitter and drift of the synthetic video faces generated by video editing.The proposed image registration algorithm is a region based image similarity measurement method,which is a coarse-to-fine refinement process.The image Pyramid is exploited to speed up the process of searching the maximum value of the similarity between two images and quickly gets the image registration at the pixel level.After the pixel level registration,the alignment algorithm is extended to sub-pixel level.The accurate location of the sub-pixel registration is calculated directly by the analytic method combined with the bilinear interpolation algorithm.After the whole image registration process completed,the faces in video sequence form a straight "pipe" on the time axis.After editing the designated area of the face image,the coordinate transformation process of face registration is inversed to restore the faces' original position back;so that there is no perceptible change in the video frames except for the edited face.In order to prevent the facial region compression or tension and to avoid the segmentation edge running into the facial feature area or facial edge in the process of video face editing,the mouth shape matching strategy is put forward.According to the relationship of the feature points near the mouth,a mouth shape index is constructed to indicate different shape of the mouth in video frames.A new source video frame sequence is formed to get the mouth shape index matched with the target,in which the movement trends of the source and target frame are coherent.An optimal mask generation method is proposed.Before Poisson image editing,the fusion area and its boundary need to be determined.In order to make the fusion result more natural or more deceptive,the optimal mask generation is proposed in this paper to avoid the fusion boundary containing non skin-color area,thus the local color bleeding problem is eliminated.The mask generation method divides the feature points into two parts,and constructs the initial mask via the feature points of the two faces.The skin model is used to constraint the boundary of blending area.It can be seen from the experimental results that the proposed method realizes the function of video face editing.Compared with other algorithms,the processed video clips are highly authentic and have strong robustness under different conditions.