An Automatic Approach for Pixel-wise Correspondence between 3D Faces

The crucial assumption of most of the model-based face analysis techniques is that any face can be generated by linear combinations of few faces. However, linear operations-Csuch as a simple pixelby- pixel addition of raw faces, are meaningless until pixel-wise correspondence of the 3D faces have been established. To overcome limitation of conventional mesh re-sampling algorithm, in this paper, we presented a fully automatic resolution, which can decrease the number of vertices and triangles while establishing correspondence between 3D faces. Based on facial features, a 3D face is segmented into 36 patches automatically at first. Each patch is then resampled to obtain a standardized patch, the same number of vertices with the template. At last, triangularly grid meshes are reconstructed for each patch individually and finally a standardized face is created by concatenating all the re-sampled patches. The experiment results show that without the complex iteration, our system is able to run quickly (resampling procedure costs less than one second) and obtain standardized 3D faces of different resolutions.