| As an important research object in the field of computer vision,face has always been the research focus of universities and research institutes.With the rapid development of computer vision technology,video face replacement,as a new technology emerging in recent years,is being valued by many universities and research institutions.Video face replacement is the replacement of the source video face to the target video face,which has an extremely important application value in video chat,privacy protection,video special effects and virtual reality.Aiming at the problem of jitter and bleeding in the particular application of 2D-based video face replacement,an algorithm of non-edge segmentation of video face without jitter and bleeding is proposed.On the basis of automatic video face replacement,the algorithm effectively prevents the phenomenon of jitter and bleeding of replacement faces.and makes the video face replacement have a more natural effect.The main contents of this paper are as follows:An algorithm for locating facial feature points in adjacent frames is proposed.Based on the active shape model,the feature points of the video face are located and the local texture of the same feature points between adjacent frames is modeled.Then,according to the similarity of local texture,the optimal matching feature points between adjacent frames are obtained,and the stability of the same feature points between adjacent frames is enhanced.The algorithm avoids the deviation of the same feature point between adjacent frames because of the random error of the algorithm,and ensures the accuracy of the subsequent video face non-edge segmentation algorithm.A spatial consistency algorithm for video faces is proposed.Procrustes Analysis method based on singular value decomposition(SVD)is used to align the source face to the target face to ensure that the size and location of the source face and the target face are consistent.The video face keeps the consistency of the space,which ensures the accuracy of the subsequent non-edge segmentation algorithm of the video face,and prevents the human face from the unnatural effect due to the different size and position of the source face and the target face.A non-edge segmentation algorithm for video faces is proposed.On the basis of ensuring the accurate location and spatial consistency of the feature points of the video face,the algorithm is based on the known feature points of the face,and uses linear interpolation to get the key points for non-edge segmentation of the face,thus the non-optimal face non-edge segmentation contour is obtained.Then,the energy function corresponding to the contour of non-edge segmentation is established.The energy function takes into account the texture smoothness of the face non-edge segmentation and the time continuity of the non-edge segmentation of adjacent frames.Finally,the energy is minimized and the optimal face contour is obtained.From the final experimental results,it can be seen that the replacement video facial expression is natural,the bleeding phenomenon is eliminated and the jitter phenomenon is obviously reduced.Therefore,the algorithm has certain theoretical and practical value for the development of video face replacement technology. |
