本論文工作的主題是圖像反卷積問題。在很多實際應用，例如生物醫學成像，地震學，天文學，遙感和光學成像中，觀測數據經常會出現令人不愉快的退化現象，這種退化一般由模糊效應(例如光學衍射限條件)和噪聲汙染(比如光子計數噪聲和讀出噪聲)造成的，這兩者都是物理儀器自身的條件限制造成的。作為一個標准的線性反問題，圖像反卷積經常被用作恢複觀測到的模糊的有噪點的圖像。我們旨在基于無偏差風險估計准則研究新的反卷積算法。本論文工作主要分為以下兩大部分。 / 首先，我們考慮在加性高斯白噪聲條件下的圖像非盲反卷積問題，即准確的點擴散函數已知。我們的研究准則是最小化均方誤差的無偏差估計，即SURE. SURE- LET方法最初被應用于圖像降噪問題。本論文工作擴展該方法至討論圖像反卷積問題.我們提出了一個新的SURE-LET算法，用于快速有效地實現圖像複原功能。具體而言，我們將反卷積過程參數化表示為有限個基本函數的線性組合，稱作LET方法。反卷積問題最終簡化為求解該線性組合的最優線性系數。由于SURE的二次項本質和線性參數化表示，求解線性系數可由求解線性方程組而得。實驗結果顯示該論文提出的方法在信噪比，圖像的視覺質量和運算時間等方面均優于其他迄今最優秀的算法。 / 論文的第二部分討論圖像盲複原中的點擴散函數估計問題。我們提出了blur-SURE -一個均方誤差修正版的無偏差估計 - 作為點擴散函數估計的最新准則，即點擴散函數由最小化這個新的目標函數獲得。然後我們利用這個估計的點擴散函數，用第一部分所提出的SURE-LET算法進行圖像的非盲複原。我們以一些典型的點擴散函數形式(高斯函數最為典型)為例詳細闡述該blur-SURE理論框架。實驗結果顯示最小化blur-SURE能夠更准確的估計點擴散函數，從而獲得更加優越的反卷積佳能。相比于圖像非盲複原，盲複原所得的圖片的視覺質量損失可忽略不計。 / 本論文所提出的基于無偏差估計的算法可擴展至其他噪聲模型。由于本論文以SURE基礎的方法在理論上並不僅限于卷積問題，該方法可用于解決數據的其他線性失真問題。 / The subject of this thesis is image deconvolution. In many real applications, e.g. biomedical imaging, seismology, astronomy, remote sensing and optical imaging, undesirable degradations by blurring effect (e.g. optical diffraction-limited condition) and noise corruption (e.g. photon-counting noise and readout noise) are inherent to any physical acquisition device. Image deconvolution, as a standard linear inverse problem, is often applied to recover the images from their blurred and noisy observations. Our interest lies in novel deconvolution algorithms based on unbiased risk estimate. This thesis is organized in two main parts as briefly summarized below. / We first consider non-blind image deconvolution with the corruption of additive white Gaussian noise (AWGN), where the point spread function (PSF) is exactly known. Our driving principle is the minimization of an unbiased estimate of mean squared error (MSE) between observed and clean data, known as "Stein's unbiased risk estimate" (SURE). The SURE-LET approach, which was originally developed for denoising, is extended to the deconvolution problem: a new SURE-LET deconvolution algorithm for fast and efficient implementation is proposed. More specifically, we parametrize the deconvolution process as a linear combination of a small number of known basic processings, which we call the linear expansion of thresholds (LET), and then minimize the SURE over the unknown linear coefficients. Due to the quadratic nature of SURE and the linear parametrization, the optimal linear weights of the combination is finally achieved by solving a linear system of equations. Experiments show that the proposed approach outperforms other state-of-the-art methods in terms of PSNR, SSIM, visual quality, as well as computation time. / The second part of this thesis is concerned with PSF estimation for blind deconvolution. We propose a "blur-SURE" - an unbiased estimate of a filtered version of MSE - as a novel criterion for estimating the PSF, from the observed image only, i.e. the PSF is identified by minimizing this new objective functional, whose validity has been theoretically verified. The blur-SURE framework is exemplified with a number of parametric forms of the PSF, most typically, the Gaussian kernel. Experiments show that the blur-SURE minimization yields highly accurate estimate of PSF parameters. We then perform non-blind deconvolution using the SURE-LET algorithm proposed in Part I, with the estimated PSF. Experiments show that the estimated PSF results in superior deconvolution performance, with a negligible quality loss, compared to the deconvolution with the exact PSF. / One may extend the algorithms based on unbiased risk estimate to other noise model. Since the SURE-based approaches does not restrict themselves to convolution operation, it is possible to extend them to other distortion scenarios. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Xue, Feng. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2013. / Includes bibliographical references (leaves 119-130). / Abstracts also in Chinese. / Dedication --- p.i / Acknowledgments --- p.iii / Abstract --- p.ix / List of Notations --- p.xi / Contents --- p.xvi / List of Figures --- p.xx / List of Tables --- p.xxii / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Motivations and objectives --- p.1 / Chapter 1.2 --- Mathematical formulation for problem statement --- p.2 / Chapter 1.3 --- Survey of non-blind deconvolution approaches --- p.2 / Chapter 1.3.1 --- Regularization --- p.2 / Chapter 1.3.2 --- Regularized inversion followed by denoising --- p.4 / Chapter 1.3.3 --- Bayesian approach --- p.4 / Chapter 1.3.4 --- Remark --- p.5 / Chapter 1.4 --- Survey of blind deconvolution approaches --- p.5 / Chapter 1.4.1 --- Non-parametric blind deconvolution --- p.5 / Chapter 1.4.2 --- Parametric blind deconvolution --- p.7 / Chapter 1.5 --- Objective assessment of the deconvolution quality --- p.8 / Chapter 1.5.1 --- Peak Signal-to-Noise Ratio (PSNR) --- p.8 / Chapter 1.5.2 --- Structural Similarity Index (SSIM) --- p.8 / Chapter 1.6 --- Thesis contributions --- p.9 / Chapter 1.6.1 --- Theoretical contributions --- p.9 / Chapter 1.6.2 --- Algorithmic contributions --- p.10 / Chapter 1.7 --- Organization --- p.11 / Chapter I --- The SURE-LET Approach to Non-blind Deconvolution --- p.13 / Chapter 2 --- The SURE-LET Framework for Deconvolution --- p.15 / Chapter 2.1 --- Motivations --- p.15 / Chapter 2.2 --- Related work --- p.15 / Chapter 2.3 --- Problem statement --- p.17 / Chapter 2.4 --- Stein's Unbiased Risk Estimate (SURE) for deconvolution --- p.17 / Chapter 2.4.1 --- Original SURE --- p.17 / Chapter 2.4.2 --- Regularized approximation of SURE --- p.18 / Chapter 2.5 --- The SURE-LET approach --- p.19 / Chapter 2.6 --- Summary --- p.20 / Chapter 3 --- Multi-Wiener SURE-LET Approach --- p.23 / Chapter 3.1 --- Problem statement --- p.23 / Chapter 3.2 --- Linear deconvolution: multi-Wiener filtering --- p.23 / Chapter 3.3 --- SURE-LET in orthonormal wavelet representation --- p.24 / Chapter 3.3.1 --- Mathematical formulation --- p.24 / Chapter 3.3.2 --- SURE minimization in orthonormal wavelet domain --- p.26 / Chapter 3.3.3 --- Computational issues --- p.27 / Chapter 3.4 --- SURE-LET approach for redundant wavelet representation --- p.30 / Chapter 3.5 --- Computational aspects --- p.32 / Chapter 3.5.1 --- Periodic boundary extensions --- p.33 / Chapter 3.5.2 --- Symmetric convolution --- p.36 / Chapter 3.5.3 --- Half-point symmetric boundary extensions --- p.36 / Chapter 3.5.4 --- Whole-point symmetric boundary extensions --- p.43 / Chapter 3.6 --- Results and discussions --- p.46 / Chapter 3.6.1 --- Experimental setting --- p.46 / Chapter 3.6.2 --- Influence of the number of Wiener lters --- p.47 / Chapter 3.6.3 --- Influence of the parameters on the deconvolution performance --- p.48 / Chapter 3.6.4 --- Influence of the boundary conditions: periodic vs symmetric --- p.52 / Chapter 3.6.5 --- Comparison with the state-of-the-art --- p.52 / Chapter 3.6.6 --- Analysis of computational complexity --- p.59 / Chapter 3.7 --- Conclusion --- p.60 / Chapter II --- The SURE-based Approach to Blind Deconvolution --- p.63 / Chapter 4 --- The Blur-SURE Framework to PSF Estimation --- p.65 / Chapter 4.1 --- Introduction --- p.65 / Chapter 4.2 --- Problem statement --- p.66 / Chapter 4.3 --- The blur-SURE framework for general linear model --- p.66 / Chapter 4.3.1 --- Blur-MSE: a modified version of MSE --- p.66 / Chapter 4.3.2 --- Blur-MSE minimization --- p.67 / Chapter 4.3.3 --- Blur-SURE: an unbiased estimate of the blur-MSE --- p.67 / Chapter 4.4 --- Application of blur-SURE framework for PSF estimation --- p.68 / Chapter 4.4.1 --- Problem statement in the context of convolution --- p.68 / Chapter 4.4.2 --- Blur-MSE minimization for PSF estimation --- p.69 / Chapter 4.4.3 --- Approximation of exact Wiener filtering --- p.70 / Chapter 4.4.4 --- Blur-SURE minimization for PSF estimation --- p.72 / Chapter 4.5 --- Concluding remarks --- p.72 / Chapter 5 --- The Blur-SURE Approach to Parametric PSF Estimation --- p.75 / Chapter 5.1 --- Introduction --- p.75 / Chapter 5.1.1 --- Overview of parametric PSF estimation --- p.75 / Chapter 5.1.2 --- Gaussian PSF as a typical example --- p.75 / Chapter 5.1.3 --- Outline of this chapter --- p.76 / Chapter 5.2 --- Parametric estimation: problem formulation --- p.77 / Chapter 5.3 --- Examples of PSF parameter estimation --- p.77 / Chapter 5.3.1 --- Gaussian kernel --- p.77 / Chapter 5.3.2 --- Non-Gaussian PSF with scaling factor s --- p.78 / Chapter 5.4 --- Minimization via the approximated function λ = λ (s) --- p.79 / Chapter 5.5 --- Results and discussions --- p.82 / Chapter 5.5.1 --- Experimental setting --- p.82 / Chapter 5.5.2 --- Non-Gaussian functions: estimation of scaling factor s --- p.83 / Chapter 5.5.3 --- Gaussian function: estimation of standard deviation s --- p.84 / Chapter 5.5.4 --- Comparison of deconvolution performance with the state-of-the-art --- p.84 / Chapter 5.5.5 --- Application to real images --- p.87 / Chapter 5.6 --- Conclusion --- p.90 / Chapter 6 --- The Blur-SURE Approach to Motion Deblurring --- p.93 / Chapter 6.1 --- Introduction --- p.93 / Chapter 6.1.1 --- Background of motion deblurring --- p.93 / Chapter 6.1.2 --- Related work: parametric estimation of motion blur --- p.93 / Chapter 6.1.3 --- Outline of this chapter --- p.94 / Chapter 6.2 --- Parametric estimation of motion blur: problem formulation --- p.94 / Chapter 6.2.1 --- Parametrized form of linear motion blur --- p.94 / Chapter 6.2.2 --- The blur-SURE framework to motion blur estimation --- p.94 / Chapter 6.3 --- An example of the blur-SURE approach to motion blur estimation --- p.95 / Chapter 6.4 --- Implementation issues --- p.96 / Chapter 6.4.1 --- Estimation of motion direction --- p.97 / Chapter 6.4.2 --- Estimation of blur length --- p.97 / Chapter 6.4.3 --- Short summary --- p.98 / Chapter 6.5 --- Results and discussions --- p.98 / Chapter 6.5.1 --- Experimental setting --- p.98 / Chapter 6.5.2 --- Estimations of blur direction and length --- p.99 / Chapter 6.5.3 --- Motion deblurring: the synthetic experiments --- p.99 / Chapter 6.5.4 --- Motion deblurring: the real experiment --- p.101 / Chapter 6.6 --- Conclusion --- p.103 / Chapter 7 --- Epilogue --- p.107 / Chapter 7.1 --- Summary --- p.107 / Chapter 7.2 --- Perspectives --- p.108 / Chapter A --- Proof --- p.109 / Chapter A.1 --- Proof of Theorem 2.1 --- p.109 / Chapter A.2 --- Proof of Eq.(2.6) in Section 2.4.2 --- p.110 / Chapter A.3 --- Proof of Eq.(3.5) in Section 3.3.1 --- p.110 / Chapter A.4 --- Proof of Theorem 3.6 --- p.112 / Chapter A.5 --- Proof of Theorem 3.12 --- p.112 / Chapter A.6 --- Derivation of noise variance in 2-D case (Section 3.5.4) --- p.114 / Chapter A.7 --- Proof of Theorem 4.1 --- p.116 / Chapter A.8 --- Proof of Theorem 4.2 --- p.116

Thesis (Ph.D.)--University of Texas at Dallas, 2007. / Includes vita. Includes bibliographical references (leaves 101-106)

Exegesis (MA--Art and Design) -- AUT University, 2007. / Cover title. Includes bibliographical references. Also held in print (136 p. : col. ill. ; 24 cm.) in the Archive at the City Campus (T 704.948 DOO)

Thesis (Ph. D.)--Florida State University, 2006. / Advisor: Anuj Srivastava, Florida State University, College of Arts and Sciences, Dept. of Statistics. Title and description from dissertation home page (viewed Sept. 22, 2006). Document formatted into pages; contains xiv, 123 pages. Includes bibliographical references.

Theses (M.S.)--Marshall University, 2006. / Title from document title page. Includes abstract. Document formatted into pages: contains ix, 99 pages. Bibliography: p. 97-99.

No description available.

Image-related communications are forming an increasingly large part of modern communications, bringing the need for efficient and effective compression. Image compression is important for effective storage and transmission of images. Many techniques have been developed in the past, including transform coding, vector quantization and neural networks. In this thesis, a novel adaptive compression technique is introduced based on adaptive rather than fixed transforms for image compression. The proposed technique is similar to Neural Network (NN)-based image compression and its superiority over other techniques is presented It is shown that the proposed algorithm results in higher image quality for a given compression ratio than existing Neural Network algorithms and that the training of this algorithm is significantly faster than the NN based algorithms. This is also compared to the JPEG in terms of Peak Signal to Noise Ratio (PSNR) for a given compression ratio and computational complexity. Advantages of this idea over JPEG are also presented in this thesis.

The enormous and ever-growing amount of web images has brought great challenges and opportunities to computer vision. A lot of recent research efforts have been devoted to the effective access of web images (e.g., web image re-ranking, content-based image retrieval), as well as to making better use of web images as auxiliary data sources. A key problem that needs to be solved is to model the semantic and visual structure of web images and to effectively organize the web image collections. This thesis proposes Visual Semantic Complex Network (VSCN), a automatically generated graph structure to model their structure and organize the web image collections. The nodes of the VSCN are compact clusters of images with visual and semantic consistency, named semantic concepts. These nodes are connected based on the visual and semantic correlations. The VSCN consists of 33,240 semantic concepts and cover 10 million images, and is expandable given more computational resources. / The VSCN enables a macroscopic understanding of the structure of web image collection. We study its structural properties from the viewpoint of complex network, and derive a great deal of valuable information. We study the connectivity and efficiency of the VSCN, and show that it is a well connected and efficient network. The study of in-degrees reflects the unequal importance of the semantic concepts. We also show the presence of community structures in the VSCN through analyzing the clustering coefficient. These studies not only gives us a better understanding of the VSCN and the web image collection at a macroscopic level, but also inspires a number of import applications of the VSCN in traditional vision tasks. / The VSCN models the structure of web images, it can be used to assist a variety of vision tasks and provide effective access to web images. We demonstrate three applications of the VSCN in this thesis. In the first application, we devise Anchor Concept Graph Distance (ACGD Distance) based on the VSCN and use this distance for web image re-ranking. As the VSCN effectively models the structures of web images, the ACG distance is able to reduce the semantic gap and better represents similarities between images. We demonstrate its effectiveness on image re-ranking benchmarks. The second application leverages the VSCN to improve the content-based image retrieval algorithms. As the relevant images on the web are connected via inter-concept correlations and grouped by community structures, we can therefore effectively reduce the search space by exploiting the structures of web images encoded in the VSCN. Therefore, the image retrieval task can greatly benefit from the VSCN. In the third application, we develop an novel image browsing scheme based on the VSCN. With the help of the VSCN structure, our proposed browsing scheme allows image browsing over the entire network via navigating the inter-concept connections. The browsing process not only bridges relevant images indexed under different textual queries, it is also guided by natural and meaningful transitions of images. We demonstrate the the VSCN-based scheme leads to better user experience for image browsing. Apart from benefiting better access to web image collections, the VSCN can be also used as an large, structured auxiliary database to assist other vision tasks. In this thesis, we present an example application of object recognition. Our method employs both imagery and textual data indexed by the VSCN, as well as its structural information. Experiments on benchmark datasets demonstrate the effectiveness of our method and the potential of the VSCN database. / 海量並且不斷增長的互聯網圖片給計算機視覺的研究帶來了很大的挑戰和機遇。許多近期的研究致力於網絡圖片的有效獲取（例如網絡圖片重排序，基於內容的圖片檢索即CBIR），以及更有效利用網絡圖片作為輔助數據源。其中一個亟待解決的關鍵問題是如何對網絡圖片間的語義和視覺結構進行建模，並據此組織這些圖片。本論文提出一個自動生成的網絡結構，即視覺語義復雜網絡（VSCN）來建模網絡圖片的結構。VSCN上的節點由一些在視覺和語義層面具有一致性的圖片聚類組成，稱之為語義概念。這些節點又由它們之間的視覺和語義相關性所連接。VSCN共包含33,240個語義概念並囊括了超過一千萬的圖片，並且可以隨著計算資源的增加而不斷增長。 / VSCN使得對網絡圖片結構的宏觀描述成為可能。我們從復雜網絡的角度對它的結構性質進行了研究，並且得出了許多有價值的信息。我們研究了VSCN的連通性和效率，說明它是一個聯通性和效率較好的網絡。對網絡入度的研究反應了語義概念具有不均等的重要性。我們通過分析聚類系數，說明了VSCN存在較明顯的社區結構。這些研究不僅僅為我們提供了對於VSCN和網絡圖片結構的宏觀理解，並且啟發了許多VSCN在計算機視覺任務中的重要應用。 / 因為VSCN建模了網絡圖片結構，它可以被用於許多計算機視覺的任務中並提供對網絡圖片有效的獲取。在本論文中我們展示了三個VSCN的應用。在第一個應用中，我們基於VSCN設計了用於網絡圖片重排序的ACG距離。由於VSCN有效地建模了網絡圖片的結構，ACG距離能夠減少語義鴻溝並更好地表達圖片之間的相似性。我們在基准數據集上展現了此方法的有效性。第二個應用利用VSCN改進了CBIR系統的。由於網絡中的相關圖片被VSCN中的語義概念間相關性所聯系，並組成了社區結構，我們可以利用此種結構信息來有效地減小搜索空間。這使得圖片檢索任務獲益，並使檢索結果得到很大改善。在第三個應用中，我們基於VSCN設計了一個新穎的圖片瀏覽模式。我們的瀏覽模式可以允許通過語義概念間的連接對VSCN進行瀏覽。這個瀏覽過程不僅聯系了屬於不同文本檢索詞的相關圖片，並且被具有指導性的圖片過渡所輔助。我們證明這種基於VSCN的瀏覽方式能夠帶來圖片瀏覽中更好的用戶體驗。除去能夠使得獲取互聯網圖片獲益之外，VSCN同樣可以被用作一個大規模的具有結構的輔助數據庫，並輔助其他計算機使用的任務。在本論文中，我們介紹了一個在物體識別的應用例。我們的方法利用的VSCN所檢索的圖片和文本信息，以及VSCN本身的結構信息。在標准數據集上的實驗證明了我們方法以及VSCN作為輔助數據庫的有效性。 / Qiu, Shi. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2014. / Includes bibliographical references (leaves 145-164). / Abstracts also in Chinese. / Title from PDF title page (viewed on 04, October, 2016). / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only.

This thesis explores the problem of how to present digital images of ancient artifacts in a manner that supports the task of visual analysis. The real object presents the "truth": exact scale, colour, and fine details. An original work of art provides the viewer with the opportunity to react directly with the object, is closest to the impact intended by the artist, and provides a tangible physical link with the past. Digital images limit and alter the experience of a work of art (1) with regard to the amount of data available (resolution), and (2) through the interpretation of the object by the producers of the digital copy (mediation). A new framework is developed to improve the understanding and presentation of virtual artifacts. This Fidelity-Mediation framework provides a continuum for considering the effects of design strategies on media used in teaching Classical arch�ology. Two small-scale experiments and follow-up interviews were undertaken to assess the usefulness of the Fidelity-Mediation framework as a descriptive model. During the experiments, quantitative analysis could detect no statistical difference in the effectiveness of different types of presentation (real object, VR object, and still digital images). This is a surprising result as it might be expected that there is nothing like seeing the real thing. Digital images provide less visual integrity. However, the digitised artifacts make up for the loss of excitement and authenticity by providing the advantage of mediated focus. Digitised artifacts thus turn out to be useful, effective study tools in the analysis of Classical art. Findings from this research are expected to generalise only to learning situations which support task orientation--situations conducive to developing personal skills and mastery--in contrast to performance orientation where the goal is to display performance relative to others. The distinction between task orientation and performance orientation is discussed in Chapter Eight of this thesis.

Thèse (Docteur ès sciences)--Université de Genève, 2002. / Summary in French. "Thèse no 3396." Includes bibliographical references.