Research On Face Replacement Algorithm Based On Video Reconstruction

With the popularity of photographic activities,image editing technology is an important branch in the computer field,and the image processing technology of human face is a hot spot of image research.Face recognition technology,facial landmark location,facial pose estimation,and other technologies has become the major of universities and institutions.At the same time face replacement technology is an emerging one,and has truly important application fields and values in privacy protection and entertainment.A face-replacement system is proposed based on video.The system replaces the performer's face in the target video footage with the user's face in the original video footage.Based on the 2D face model,the whole replacement process does not require additional control of expression database.The main processing module is divided into five parts,namely,facial key point location module,facial expression matching module,facial skin color conversion module,face deformation module and face fusion module.This paper has the following three innovations:A facial expression matching strategy is proposed.Original video clip is reconstructed according to the change of actor's expression in the target video clip based on the new matching rules for joint of expression and motion.After the matching,the facial expression in the original video clip and the facial expression in the target video clip have the same facial pose consistency and the facial motion consistency.This paper proposes a skin color transfer algorithm.According to the location of the feature points,the area of the skin color transform is specified.In the YCbCr color space,the skin color region is acquired for the first time to extract the skin color data.Using the distribution characteristics of skin color components in Lab space,the accurate extraction of pure skin color information is acquired.The conversion of skin color is achieved through the corresponding conversion rules.After skin color transformation,the skin color of the performer in the target video clip is consistent with that of the original video clip,which improves the result of final face fusion.And proposes the strategy of face deformation from local to global.Given the location of the known feature points and video clip after face alignment,the deformation of human face from local to globe is realized by the method of the triangulation algorithm and the affine transformation method.After the face reshaped,the key parts of the face in the target video frames are erased,and then the original feature points are used to generate optimal fusion boundary.Finally,the Poisson editing technique is used to achieve the fusion of human faces.The strategy of the optimal face fusion improves the naturalness of the face fusion result.From the experimental results,it can be seen that the replacement video clips looks natural,and the whole process need no additional human-computer interaction,which reduces the operation complexity of the system and shows the practicability of the system.The face replacement system does not depend on the 3D human face model,and does not require the facial expression similarity between the original video footage and the target video.It has advantages of simplicity of processing and robustness of the application.The system has a certain theoretical and practical value for the development of face replacement technology,especially the replacement of moving human faces in video.