An improved multi-view face synthesis method based on point distribution models

A multi-view face synthesis method has been proposed based on point distribution models (PDMs) for face recognition across pose, which strongly relies on the “pose parameters” in the PDM to control the up-down and/or left-right rotations of faces. However, the pose parameters would not appear in the fixed positions, and that may result in undesired virtual faces with some definite training data. Thus, a simple yet effective automatic inspection strategy is introduced in this paper to find the corresponding pose parameters for the desired face pose transformation, which could be used to improve the performance of the existing face synthesis methods. Meanwhile, a counterexample is given to illustrate show the latent defect of the existing methods. Experimental results are provided to validate the effectiveness of the proposed solution.