基于样本库的人脸图像处理方法

Human Face Super-Resolution Method Based on Train-Database

本文给出了一种基于样本库的人脸图像超分辨率方法,这一方法结合了人脸图像全局相似性和局部特征相似性的约束.局部结构的相似性将超分辨率解约束到人脸局部结构空间,不同尺度上的局部相似性指导搜索最优解,然后将局部结构重建的人脸图像投影到由人脸图像确定的特征人脸空间.实验结果表明该算法得到的人脸具有很好的准确性和视觉效果.

In this paper, we propose a human face super-resolution method based on training. This method composes similarity constrains of human face both in global and local manner. The local structure of human face ensures the solution of super-resolution to be in the space characterized by human face local structure. And the local structure in different scales instruct the searching to the optimal solution, then project the reconstructed face by local structure to the eigenface space determined by given human face database. Experiments show that the results have very good accuracy and high visual quality.