本文为爱尔兰都柏林大学（作者：Claire Gallagher）的博士论文，共225页。
本文主要研究基于示例的图像处理方法。基于示例的图像处理是对于任何类别的图像处理操作都适用的通用术语，其中所讨论的图像操作和分析由一组示例图像引导。本文主要研究两种应用，分别是纹理合成和图像分割，同时提出了基于示例的图像处理。
给定一个纹理示例，一种成功的纹理合成算法的目标是产生在感知上类似于样本纹理的新纹理。在这个过程中的主要挑战之一是示例纹理的建模。以往的工作已经表明，基于隐式模型的算法比刚性的基于显式模型的算法更成功。在此基础上，结合隐式建模技术和基于小波的图像分析的优点，提出了一种新的纹理合成算法。双树复小波变换具有良好的方向选择性和移位不变性，这两种特性都使得该算法更适合于纹理分析。新提出的算法是在小波域中进行合成，其结果是一个尺度无关的有效过程，它具有足够的鲁棒性以适应广泛的纹理分析要求。
在纹理合成算法的基础上，重点对图像进行分割。任何图像分割处理的目标是给观察图像中的每个像素分配一个标签，指示出该像素属于哪个区域或类别。全自动或无监督分割是一个病态问题，所以为了进行约束限制求解，将一个与分割相似的示例图像集作为算法输入。该示例图像集已经被预先分割，因此可以用于指导分割过程。这种类型的半自动分割可以被看作是分割和物体识别的交织。作为基于示例的图像处理的工作的一部分，已经开发了一种新的分割算法，这种基于示例的分割算法是以与合成过程相同的隐式建模技术为基础的。然而，为了使求解有效，将观测图像的隐式建模与标签字段的显式建模进行了结合。贝叶斯框架提供了这种并行建模技术的自然表达，在该框架下提出了新的算法，并给出了一些样本图像的分割结果。
This thesis is concerned with example based image processing. Examplebased image processing is a general term for any class of image processingoperation where the manipulation and analysis of the image in question isguided by some set of example images. This thesis focuses on two applications,texture synthesis and image segmentation, in which example based image processingis proposed.
Given an example texture, the goal of a successfultexture synthesis algorithm is to generate new texture which is perceptuallysimilar to the sample texture. One of the main challenges in this process isthe modelling of the example texture. Previous work has shown that those algorithmsbased on implicit modelling are more successful than those based on the morerigid explicit models. Based on this observation a new texture synthesis algorithmis developed which combines the strength of the implicit modelling techniquewith wavelet based image analysis. The Dual-Tree Complex Wavelet Transform usedin this work has associated with it good directional selectivity and shiftinvariance. Both of these properties make it well suited for texture analysis.The new synthesis algorithm exploits this by performing synthesis in the waveletdomain. The result is a scale independent efficient process which is robustenough to work for a wide range of textures. Building on the strength of thistexture synthesis algorithm, the focus then turns to image segmentation. Thegoal of any image segmentation process is to assign to each pixel in an observedimage a label indicating to which region or class that pixel belongs. Fullyautomated or unsupervised segmentation is an ill-posed problem and so in orderto constrain the solution an example image set whose content is similar to thatto be segmented is given as an input. This example image set has been segmentedapriori and so can be used to guide the segmentation process. This type ofsemi-automated segmentation can be viewed as the interleaving of segmentationand object recognition. As part of this work on example based processing, a newsegmentation algorithm has been developed. This example based segmentationalgorithm is based on the same implicit modelling technique as the synthesisprocess. However, in order to regularise the solution, implicit modelling ofthe observed image is combined with an explicit modelling of the label field.The Bayesian framework provides a natural expression for such parallelmodelling techniques. The new algorithm is presented under this framework andsome sample segmentation results are given.
下载英文原文地址：
http://page2.dfpan.com/fs/2lcej23218291162361/
更多精彩文章请关注微信号：
