引言部分的结构与语言特点
王标 S150101132
第一篇（Cascades of Regression Tree Fieldsfor Image Restoration）
（1） 句1-3，33-36提供背景信息，回顾前期相关文献，说明图像恢复是一个
热门的研究领域；句4-6，37-39通报目前研究现状，讨论当前使用先验知识来处理图像，只是盲目的去模糊；句7-10，40-42提出当前研究存在的问题，提出当前的研究存在一些特定参数和模糊内核的问题，盲目的去模糊存在很大的问题；句11-29陈述研究课题的目的，由当前研究的不足，作者提出用回归树的级联来处理图像。可以扩展到非均匀模糊图像；句30-32介绍科研论文的结构，本文从六个方面来介绍作者所提出的回归树处理图像的方法。
（2） lack of，are necessary to
We argue that the lack of discriminative methods for these applications stems from their more challenging data term with additional instance-specific parameters, which are necessary to capture the image corruption process properly. Difficult
In a discriminative approach it is, however, quite difficult to copewith such instance-specific parameters. relatively little
However, relatively little attention has been paid to non-blind deblurring, that is, restoring the image given known or estimated image blur. this is an important problem
Yet, this is an important problem since most blind deblurring approaches separate the problem into blur estimation and non-blind deblurring (theoretically justified by Levin et al. [13]).
（3）be expressed，arbitrary In this paper we introduce a discriminative image restoration approach for applications
that can be expressed via arbitrary quadratic data terms (Gaussian likelihoods). Address
We address the challenge of capturing the input distribution variability by using a semi-parametric approach Introduce
Motivated by that, we introduce a model cascade based on regression tree fields. Predicts ，useful ，To overcome this limitation，it is important
第二篇（Image Restoration Using Gaussian Mixture ModelsWith Spatially Constrained Patch Clustering）
（1） 句1-5提供背景信息，回顾前期相关文献，说明现在的图像恢复是用
y = Hx + v，由y得到x；句6-14通报目前研究现状，讨论patch-based image restoration 等方法在图像恢复的作用；句15-17提出当前研究存在的问题，提出当前的图像恢复的方法对于非线性图像的恢复是存在误差的；句18-41陈述研究课题的目的，陈述了GMM在图像处理方面的应用的优点，作者将用公式和实验来解决图像去噪和图像插值等问题；句30-32介绍科研论文的结构，本文从四个方面来介绍作者所提出的globalGMM处理图像的方法。
（2） the strict conditions drawback
However, the strict conditions for obtaining the exact sparse representations in the sparsity promoting methods [13]–[15], can be a drawback of these nonlinear methods for the image restoration. More noticeable
This disadvantage is more noticeable when the degrading operator H is not the identity matrix. Harder
（3） propose
we propose a model that uses a same multivariate Gaussian probability distribution for similar image patches in a neighborhood. use the idea of
assume that is used to more likely
outperforms is successful alreadyproposed reported solving more detailed
引言作业
