A facial texture generation method using curvature-reflected size-variable block matching

Abstract

Generating natural facial textures is a significant technique in 3D human face modeling. In this paper, we propose an ellipsoidal face model-based image registration method to generate realistic facial textures. The suggested algorithm registers multiple 2D face images captured in different views with size-variable block matching that reflects both left-and-right and up-and-down face curvatures. For accurate block matching, the algorithm simultaneously considers color and texture features. It then employs weighting factor-based mosaiking and ghosting artifact elimination techniques to generate natural facial textures from registered images. Experimental results are given to demonstrate that the suggested ellipsoidal face model-based image registration algorithm can perform more effectively than other conventional registration approaches. © 2013 International Information Institute.