一种新颖的人脸图像超分辨率技术研究

本文针对基于学习的人脸图像超分辨率技术进行研究，采用多分辨率的幻觉脸算法。针对Baker方法（文献[1]）建立的图像金字塔提取高频细节不够丰富的缺点，采用Kirsch算子提取高频特征。它与Baker的一阶、二阶灰度算子结合，能够提取更多的图像信息，使得匹配更为准确。另外将流形学习中的LLE算法思想引入匹配复原过程中，使复原结果获取更完备的高频信息。实验结果表明，本文预测得到的先验模型更为准确，使得最终复原的人脸图像具有更好的视觉效果。

The Novel method of Face Hallucination

This paper focuses on study of super-resolution image of face(Face Hallucination). A novel face hallucination algorithm is proposed in this paper based on multi-resolution pyramid structure. Different from Baker method (reference [1]), We add Kirsch operator to get high-frequency feature of image. Kirsch feature combined with the first and second order gray feature can extract more image information, make the matching process more accurate. In addition, we introduce one of the manifold learning methods, locally linear embedding (LLE) for matching process, to make priori model more accurate, more comprehensive, and having more higher frequency information. Experimental results show that this model predicted a priori be more accurate, it will make the recovery of facial images with better visual effect.