基于纹理合成的图像修复研究

摘要图像修复是图像处理领域中的一个重要内容，图像修复的目的即是根据图像现有的信息来自动恢复丢失的信息。传统的图像修复方法思路分为两类，一类是偏微分方程（Partial Differential Equations ）的方法，该方法着眼于图像中的结构性信息，另一类是纹理合成（texture synthesis）的方法，该方法着眼于图像中的纹理信息。60131

在2004年，微软研究院的criminisi博士在IEEE上发表了关于图像处理的论文，提出了基于样例的图像修复算法，在学术领域被称作criminisi算法。该算法从分类上属于纹理合成法，但是它借鉴了PDE方法对结构性信息的考虑，因此在修复过程中同时关注到图像的结构性信息和纹理性信息，使得算法在图像的处理质量、处理时间、处理稳定性这三者的综合表现上得到了良好的评价。

本文是对criminisi算法的具体实现。首先对图像和图像修复技术进行简要的介绍；接下来简要阐述criminisi算法的原理；然后详细说明在matlab2012工作平台下对criminisi算法的具体实现，其中对修复优先级的具体实现采用了一种独特的方法；最后是通过实际处理的结果对程序和算法进行分析，并得到相关结论。

毕业论文关键字图像修复纹理合成 criminisi

毕业设计说明书（论文）外文摘要

Title Image restoration based on texture synthesis

Abstract Image inpanting is an important research topic in the area of image restoration. Its objective is to restore the lost information according to the total information of a whole image. There are two kinds of classical inpainting techniques: one is image inpainting based on the geometric image models, which mainly focuses on the constructive part of the image; the other is image completion based on texture synthesis,which mainly focuses on the textural part of the image.

In 2004, Criminisi from the Microsoft Research published a piece of paper on IEEE, proposing an advanced algorithm on image inpanting. This algorithm takes example by image completion based on texture synthesis. But its main features come from image inpainting based on the geometric image models. By paying especially attention to the constructive part of the image as image inpainting, Criminisi introduced restoration priority to his algorithm, and this attempt has been proved effective and efficient in image restoration in academia.

This paper attempts to realize criminisi algorithm. First, it gives a brief introduction of the image and digital image processing. Then this paper describes image restoration by criminisi algorithm. In the third part, it expounds the concrete realization of criminisi algorithm in detail. At the end, a number of examples are demonstrated to make some further discussions, and then get the conclusion.

Keywords image restoration, texture synthesis, criminisi

1 绪论 1

1.1 图像与数字图像处理基础 1

1.2 数字图像的描述 1

1.3 数字图像修复基础 1

1.4 关于matlab 1

2 Criminisi算法 1

2.1 两类经典算法 1

2.2 纹理合成方法的表述 1

2.3 criminisi算法的核心思想 2

3 Criminisi算法的具体实现 2

3.1 初始化部分 2

3.1.1 读入图像与创建辅助变量基于纹理合成的图像修复研究:http://www.751com.cn/jisuanji/lunwen_65507.html