人脸图像超分辨率的自适应流形学习方法  

Adaptive Manifold Learning Method for Face Hallucination

作　　者：张雪松[1] 江静[2] 彭思龙[1] 

机构地区：[1]中国科学院自动化研究所国家专用集成电路设计工程技术研究中心,北京100190 [2]华北科技学院机械与电气工程系,北京101601

出　　处：《计算机辅助设计与图形学学报》2008年第7期856-863,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家科技支撑计划重点项目(2006BAK07B04)

摘　　要：样本规模与使用方法是基于学习的超分辨率中的一个重要问题.面向人脸图像超分辨率重建,提出一种基于局部保持投影(LPP)的自适应流形学习方法.由于能够揭示隐含在高维图像空间中的非线性结构,LPP是一种可以在局部人脸流形上分析其内在特征的、有效的流形学习方法.通过在LPP特征子空间中动态搜索出与输入图像块最相似的像素块集合作为学习样本,实现了自适应样本选择,并且利用动态样本集合通过基于像素块的特征变换方法有效地恢复出低分辨率人脸图像中缺失的高频成分.实验结果证实:通过在局部人脸流形上自适应地选择学习样本,文中方法可以仅使用相对少量的样本来获得很好的超分辨率重建结果.The size of training set as well as the usage thereof is an important issue of learning-based super-resolution. This work presents an adaptive learning method for face hallucination using Locality Preserving Proiection （LPP）. LPP is an efficient manifold learning method that can be used to analyze the local intrinsic features on the manifold of local facial areas by virtue of its ability to reveal nonlinear structures hidden in the high-dimensional image space. We fulfilled the adaptive sample selection by searching out patches online in the LPP sub-space, which makes the resultant training set tailored to the testing patch, and then effectively restored the lost high-frequency components of the lowresolution face image by patched-based eigen transformation using the dynamic training set. The experimental results fully demonstrate that the proposed method can achieve good super-resolution reconstruction performance by utilizing a relative small amount of samples.

关 键 词：人脸图像 超分辨率 局部保持投影 流形学习 非监督学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]