相似图像块逼近的人脸超分重建残差补偿框架

为了提高人脸图像超分辨率重建算法中残差补偿步骤的效果，提出一种通用的基于内容相似图像块线性组合逼近的残差补偿框架，不经过搜索步骤，使用训练集人脸图像同一内容的图像块来进行运算。所提框架中的全局重建步骤，可以使用不同的重建方法。实验结果表明，在这种框架下的残差补偿方法，相比经典的邻域嵌入残差补偿方法，可以更好地恢复出初步重建的人脸图像细节信息。因为这是一种通用的残差补偿方法，从而可以推测凡使用邻域残差补偿的算法，均可借助本算法框架将重建结果进一步的提升。

Residue compensation framework for human face super-resolution reconstruction based on linear combination approximation of similarity image patches

A general residue compensation framework based on similar content and linear combination of image patches is proposed to improve effect of residue compensation step in the human face super-resolution reconstruction algorithm.With the framework,the patches with the same content in the training set human face images is used to compute without searching step.The different reconsrtruction method is allowed in the global reconstruction step of the proposed framework.The experimental results indicate that the proposed residue compensation method can recover the detail of preliminarily-reconstructed human face image much better than the typical neighbourhood embedded residue compensation methods.Since it is a general residue compensation framework,it can derive any algorithms which adopt the neighbourhood residue compensation,and improve the reconstruction result by means of this algorithm framwork.