Face Fusion Based On Key Points Detection

"Photo Sharing" occupies a large part of the social scenes.QQ Space,Weibo and Wechat occupy a large number of pages in the circle of friends.At present,most applications are based on face recognition technology.With the continuous improvement of face recognition rate,more and more researchers have started to dig deep into the local information of face and put forward more interesting and valuable applications."Changing face" is actually one of them.Face and face change is actually a face synthesis technology,through the detection of key points to generate mask templates,the use of cutout,map the way to achieve the exchange of two face areas,and do some image fusion to face and around The environment is very good convergence.For this technology,its integration is still based on the region extraction,color conversion,edge fusion.The concept of face fusion proposed by this subject stems from the improvement of face synthesis technology.The difference between face fusion and human face synthesis is that the form of this fusion is aiming at the fusion of more detailed shapes and textures.All objects have an object,so we can implement this object through the local characteristics of the object object.Principle and image deformable technology to face fusion technology to achieve a certain degree of innovation in the application of the main contents of this paper are as follows.:First of all,in the ROI detection of human face,a binocular detection method based on two-level positioning frame is proposed to realize the precise positioning of both eyes.And according to the positioning results of both eyes to solve the problem of input image size and location to facilitate post-fusion processing.Then,in the extraction of the key points of the face,several classical key point localization algorithms and their improved algorithms are analyzed and compared.Finally,the STASM algorithm is selected as the localization algorithm and applied to the face fusion technology.In the process of locating the key points,this paper takes into account the needs of the subject,and adds the key points at the forehead based on the STASM algorithm to adjust the source images that need to be fused.In the face triangle mesh model generation part,this paper applies the improved Delaunay meshing algorithm,that is,the merging algorithm of divide-and-convolution method and point-by-point interpolation method to generate the triangle mesh of face quickly.Finally,image warping and face fusion,this paper applies the method of affine transformation and bilinear interpolation to image fusion of the generated face fusion grid sequences.In the process of fusion,the method of "reverse mapping" and the approach of the second order linear interpolation are used to avoid the "empty" and "ghost" phenomenon of generating the target image and effectively avoid the situation of discontinuous pixels and improve Image quality of fusion result.Finally,the experimental results show that the proposed face fusion based on the detection of key points is feasible overall,and it can achieve a good fusion efect on the richness of the face images with rich texture.