An ellipsoidal model for generating realistic 3D facial textures

摘要

In this paper, we suggest an image registration algorithm based on an ellipsoidal model with size-variable blocks that is similar in shape to a human face. While blocks are matched, the existing cylindrical model only considers that left-right curvature can accomplish a correct alignment on left and right images. However, registration errors are produced from up and down images as the cylindrical model does not reflect human head and jaw shape characteristics. The proposed algorithm exploits a block matching algorithm that uses size-variable blocks. The left-right and up-down curvature of the ellipsoidal face model is considered and input images are correctly aligned using their correlation. Next, we employ an image mosaic technique to generate a realistic facial texture from the aligned images. We stitch the images by assigning linear weights according to the overlapped regions and remove ghost effects to make the facial texture more natural. The experimental results show that the proposed algorithm can realistically generate 3D facial textures compared to other conventional methods.