小样本训练集下人脸图像超分辨率重构算法  

Super-resolution reconstruction of face images for small sample set

作　　者：陈文[1] 黄东军[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《企业技术开发》2010年第3期7-9,共3页Technological Development of Enterprise

基　　金：国家自然科学基金项目[语义马尔可夫网络下人脸图像非线性超分辨率算法(编号:60873188)]

摘　　要：文章提出了一种针对小样本训练集的人脸图像超分辨率重构算法。该算法在马尔可夫网络模型的基础上,将视觉相容性检查准则引入到候选分块的匹配准则中,提高了匹配精度和块间兼容性。将权值机制融合进图像的重构过程,强化了训练集的有用性。实验表明,在小样本训练集下,算法能有效地抑制块效应和局部失真现象,明显改善超分辨率图像的质量。This paper proposed a super-resolution algorithm of face images based on small sample set. Based on Markov Network, this set puts visual inspection criteria to the compatibility of the candidate matching block, our method improves the efficiency and block matching degree.Putting the weight mechanism melted into the image reconstruction processing can expand the usefulness of the training set. Experimental results demonstrate that the algorithm can effectively prevent the blocking effect and have a better super-resolution image quality.

关 键 词：人脸图像 超分辨率 小样本 视觉相容性 权值 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]