內容簡介
Image Processing and Analysis: Variational, PDE, Wavelet, andStochastic Methods is systematic and well organized, The authorsfirst investigate the geometric, functional, and atomic structures ofimages and then rigorously develop and analyze several imageprocessors. The book is comprehensive and integrative, covering thefour most powerful classes of mathematical tools in contemporaryimage analysis and processing while exploring their intrinsicconnections and integration. The material is balanced in theory andcomputation, following a solid theoretical analysis of model buildingand performance with computational implementation and numerical examples.
This book is written for graduate students and researchers inapplied mathematics, computer science, electrical engineering, andother disciplines who are interested in problems in imaging andcomputer vision. It can be used as a reference by scientists withspecific tasks in image processing, as well as by researchers with ageneral interest in finding out about the latest advances.
內頁插圖
目錄
List of Figures
Preface
1 Introduction
1.1 Dawning of the Era of Imaging Sciences
1.1.1 Image Acquisition
1.1.2 Image Processing
1.1.3 Image Interpretation and Visual Intelligence
1.2 Image Processing by Examples
1.2.1 Image Contrast Enhancement
1.2.2 Image Denoisirg
1.2.3 Image Deblurring
1.2.4 Image Inpainting
1.2.5 Image Segmentation
1.3 An Overview of Methodologies in Image Processing
1.3.1 Morphological Approach
1.3.2 Fourier and Spectral Analysis
1.3.3 Wavelet and Space-Scale Analysis
1.3.4 Stochastic Modeling
1.3.5 Variaticnal Methods
1.3.6 Partial Differential Equations(PDEs)
1.3.7 Different Approaches Are Intrinsically Interconnected
1.4 Organization of the Book
1.5 How to Read the Bcok
2 Some Modern Image Analysis Tools
2.1 Geometry of Curves and Surfaces
2.1.I Geometry of Curves
2.1.2 Geometry of Surfaces in Three Dimensions
2.1.3 Hausdorff Measures and Dimensions
2.2 Functions with Bounded Variations
2.2.1 Total Variatien as a Radon Measure
2.2.2 Basic Properties of BV Functions
2.2.3 The Co-Area Formula
2.3 Elements of Thermodynamics and Statistical Mechanics
2.3.1 Essentials of Thermodynamics
2.3.2 Entropy and Potentials
2.3.3 Statistical Mechanics of Ensembles
2.4 Bayesian Statistical Inference
2.4.1 Image Processing or Visual Perception as Inference
2.4.2 Bayesian Inference: Bias Due to Prior Knowledge
2.4.3 Bayesian Method in Image Processing
2.5 Linear and Nonlinear Filtering and Diffusion
2.5.1 Point Spreading and Markov Transition
2.5.2 Linear Filtering and Diffusion
2.5.3 Nonlinear Filtering and Diffusion
2.6 Wavelets and Multiresolution Analysis
2.6.1 Quest for New Image Analysis Tools
2.6.2 Early Edge Theory and Marr’s Wavelets
2.6.3 Windowed Frequency Analysis and Gabor Wavelets
2.6.4 Frequency-Window Coupling: Malvar-Wilson Wavelets
2.6.5 The Framework of Multiresolution Analysis (MRA)
2.6.6 Fast Image Analysis and Synthesis via Filter Banks
3 Image Modeling and Representation
3.1 Modeling and Representation: What, Why, and How
3.2 Deterministic Image Models
3.2.1 Images as Distributions (Generalized Functions)
3.2.2 Lp Images
3.2.3 Sobolev Images Hn(Ω)
3.2.4 BV Images
3.3 Wavelets and Multiscale Representation
3.3.1 Construction of 2-D Wavelets
3.3.2 Wavelet Responses to Typical Image Features
3.3.3 Besov Images and Sparse Wavelet Representation
3.4 Lattice and Random Field Representation
3.4.1 Natural Images of Mother Nature
3.4.2 Images as Ensembles and Distributions
3.4.3 Images as Gibbs’ Ensembles
3.4.4 Images as Markov Random Fields
3.4.5 Visual Filters and Filter Banks
3.4.6 Entropy-Based Learning of Image Patterns
3.5 Level-Set Representation
3.5.1 Classical Level Sets
3.5.2 Cumulative Level Sets
3.5.3 Level-Set Synthesis
3.5.4 An Example: Level Sets of Piecewise Constant Images
3.5.5 High Order Regularity of Level Sets
3.5.6 Statistics of Level Sets of Natural Images
3.6 The Mumford-Shah Free Boundary Image Model
3.6.1 Piecewise Constant 1-D Images: Analysis and Synthesis
3.6.2 Piecewise Smooth 1-D Images: First Order Representation
3.6.3 Piecewise Smooth I-D Images: Poisson Representation
3.6.4 Piecewise Smooth 2-D Images
3.6.5 The Mumford-Shah Model
3.6.6 The Role of Special B V Images
4 Image Denoising
4.1 Noise: Origins. Physics. and Models
4.l. 1 Origins and Physics of Noise
4.1.2 A Brief Overview of 1-D Stochastic Signals
4.1.3 Stochastic Models of Noises
4.1.4 Analog White Noises as Random Generalized Functions
4.1.5 Random Signals from Stochastic Differential Equations
4.1.6 2-D Stochastic Spatial Signals: Random Fields
4.2 Linear Denoising: Lowpass Filtering
4.2.1 Signal vs. Noise
4.2.2 Denoising via Linear Filters and Diffusion
4.3 Data-Driven Optimal Filtering: Wiener Filters
4.4 Wavelet Shrinkage Denoising
4.4.1 Shrinkage: Quasi-statistical Estimation of Singletons
4.4.2 Shrinkage: Variational Estimation of Singletons
4.4.3 Denoising via Shrinking Noisy Wavelet Components
4.4.4 Variational Denoising of Noisy Besov Images
4.5 Variational Denoising Based on BV Image Model
4.5.1 TV. Robust Statistics. and Median
4.5.2 The Role of TV and BV Image Model
4.5.3 Biased Iterated Median Filtering
4.5.4 Rudin. Osher. and Fatemi's TV Denoising Model
4.5.5 Computational Approaches to TV Denoising
4.5.6 Duality for the TV Denoising Model
4.5.7 Solution Structures of the TV Denoising Model
4.6 Denoising via Nonlinear Diffusion and Scale-Space Theory
4.6.1 Perona and Malik's Nonlinear Diffusion Model
4.6.2 Axiomatic Scale-Space Theory
4.7 Denoising Salt-and-Pepper Noise
4.8 Multichannel TV Denoising
4.8.1 Variational TV Denoising of Multichannel Images
4.8.2 Three Versions of TV[u]
5 Image Deblurring
5.1 Blur: Physical Origins and Mathematical Models
5.1.1 Physical Origins
5.1.2 Mathematical Models of Blurs
5.1.3 Linear vs. Nonlinear Blurs
5.2 Ill-posedness and Regularization
5.3 Deblurring with Wiener Filters
5.3.1 Intuition on Filter-Based Deblurring
5.3.2 Wiener Filtering
5.4 Deblurring of BV Images with Known PSF
5.4.1 The Variational Model
5.4.2 Existence and Uniqueness
5.4.3 Computation
5.5 Variational Blind Deblurring with Unknown PSF
5.5.1 Parametric Blind Deblurring
5.5.2 Parametric-Field-Based Blind Deblurring
5.5.3 Nonparametric Blind Deblurring
6 Image Inpainting
6.1 A Brief Review on Classical Interpolation Schemes
6.1.1 Polynomial Interpolation
6.1.2 Trigonometric Polynomial Interpolation
6.1.3 Spline Interpolation
6.1.4 Shannon's Sampling Theorem
6.1.5 Radial Basis Functions and Thin-Plate Splines
6.2 Challenges and Guidelines for 2-D Image Inpainting
6.2.1 Main Challenges for Image Inpainting
6.2.2 General Guidelines for Image Inpainting
6.3 Inpainting of Sobolev Images: Green's Formulae
6.4 Geometric Modeling of Curves and Images
6.4.1 Geometric Curve Models
6.4.2 2-. 3-Point Accumulative Energies. Length. and Curvature.
6.4.3 Image Models via Functionalizing Curve Models
6.4.4 Image Models with Embedded Edge Models
6.5 Inpainting BV Images (via the TV Radon Measure)
6.5.1 Formulation of the TV Inpainting Model
6.5.2 Justification of TV Inpainting by Visual Perception
6.5.3 Computation of TV lnpainting
6.5.4 Digital Zooming Based on TV Inpainting
6.5.5 Edge-Based Image Coding via Inpainting
6.5.6 More Examples and Applications of TV Inpainting
6.6 Error Analysis for Image Inpainting
6.7 Inpainting Piecewise Smooth Images via Mumford and Shah
6.8 Image Inpainting via Euler's Elasticas and Curvatures
6.8.1 Inpainting Based on the Elastica Image Model
6.8.2 Inpainting via Mumford-Shah-Euler Image Model
6.9 Inpainting of Meyer's Texture
6.10 Image Inpainting with Missing Wavelet Coefficients
6.11 PDE Inpainting: Transport. Diffusion. and Navier-Stokes
6.11.1 Second Order Interpolation Models
6.11.2 A Third Order PDE Inpainting Model and Navier-Stokes
……
7 Image Segmentation
Bibliography
Index
前言/序言
《圖像處理與分析:變分、PDE、小波及隨機方法》 圖書簡介 本捲,作為“國外數學名著係列(續一 影印版)”中的第38部,隆重推齣《圖像處理與分析:變分、PDE、小波及隨機方法》,為廣大數學、計算機科學、工程學以及相關領域的科研人員、研究生和高級工程師提供瞭一部集理論深度與應用廣度於一體的權威著作。本書聚焦於現代圖像處理與分析領域中最具影響力和活力的幾個核心數學工具和方法,即變分方法、偏微分方程(PDE)、小波理論以及隨機方法。通過係統深入地闡述這些數學工具在圖像領域的獨特作用和強大能力,本書旨在構建一個堅實的理論框架,並引導讀者掌握解決復雜圖像問題的關鍵技術。 本書的編寫宗旨在於 bridging the gap between abstract mathematical theory and practical image processing challenges. 圖像,作為我們感知世界最直觀的方式,其背後蘊含著海量的信息。如何從原始圖像中提取、理解、增強、恢復、分割,乃至理解其內容,是信息時代的核心課題之一。而近年來,數學領域的進步,尤其是上述幾個關鍵分支的發展,為解決這些挑戰提供瞭前所未有的強大工具。本書正是係統梳理和呈現這些數學工具如何被巧妙地應用於圖像處理與分析的最新成果。 核心內容概覽: 1. 變分方法 (Variational Methods): 變分方法是優化問題的一個強大框架,它通過最小化一個能量泛函來尋找最優解。在圖像處理中,許多問題都可以轉化為尋找一個能量最小化的圖像。本書將詳細介紹變分法的基本原理,包括歐拉-拉格朗日方程、正則化理論等。重點將放在如何為圖像處理問題構建閤適的能量函數,例如: 圖像去噪 (Image Denoising): 傳統的圖像去噪方法往往會模糊圖像細節,而變分方法能夠通過正則化項來平衡平滑度和細節保持,例如Total Variation (TV) 模型,它在保持圖像邊緣的同時有效去除噪聲。本書將深入探討TV模型的數學理論,以及其在不同噪聲模型下的變種和改進。 圖像恢復 (Image Restoration): 圖像模糊、缺失等問題可以通過變分方法進行建模和求解。例如,使用變分模型來反捲積模糊圖像,或者修復缺失的圖像區域。 圖像分割 (Image Segmentation): 將圖像劃分為不同的區域是圖像分析的關鍵步驟。Chan-Vese 模型等基於變分原理的水平集方法,能夠根據圖像的灰度、紋理等特徵自動分割齣目標物體,即使目標邊界不連續或存在噪聲。本書將深入講解這些模型背後的數學推導和算法實現。 2. 偏微分方程 (Partial Differential Equations - PDE): 偏微分方程在描述連續介質的演化規律方麵扮演著至關重要的角色,而圖像本身就可以被視為一種二維的連續信號。因此,PDE在圖像處理中展現齣瞭強大的生命力。本書將係統介紹PDE在圖像領域的應用,包括: 圖像擴散 (Image Diffusion): 類似於熱傳導方程,圖像擴散模型可以用於平滑圖像,去除噪聲,同時可以選擇性地保留邊緣。Anisotropic diffusion (各嚮異性擴散) 模型,例如Perona-Malik模型,能夠根據圖像局部特徵決定擴散的方嚮和強度,從而實現選擇性平滑,在去噪和特徵提取方麵效果顯著。 圖像分割與重建: PDE可以用來構建演化方程,驅動圖像分割過程。例如,基於麯率流的幾何活動輪廓模型(如Mean Curvature Flow),它們通過演化麯綫的幾何性質來逼近物體的邊界。 運動估計與光流: PDE方法也被用於計算圖像序列中的運動信息,例如光流的估計。 3. 小波理論 (Wavelet Theory): 小波變換是一種時頻局部化變換,相比於傅裏葉變換,它能夠同時捕捉信號的頻率信息和局部位置信息,這對於分析具有奇異點或不連續特徵的圖像至關重要。本書將深入探討小波理論在圖像處理中的應用: 圖像壓縮 (Image Compression): 小波變換能夠有效地將圖像能量集中在少數幾個小波係數上,從而實現高效的圖像壓縮,例如JPEG2000標準就采用瞭小波變換。 圖像去噪: 在小波域中,噪聲的係數通常分布在幅度較小的區域,而圖像的有用信息則集中在較大的係數上。通過對小波係數進行閾值處理,可以有效地去除噪聲,同時保留圖像的細節。本書將介紹不同的小波去噪算法,如軟閾值和硬閾值方法。 圖像融閤 (Image Fusion): 將來自不同傳感器或不同處理方式的圖像融閤,以獲得更豐富的信息。小波變換能夠從不同尺度和方嚮上分析圖像,並進行多分辨率的融閤。 多分辨率分析: 小波提供瞭一種強大的多分辨率分析框架,使得我們可以從不同的尺度上觀察和處理圖像,這對於分析圖像的結構和紋理非常有用。 4. 隨機方法 (Stochastic Methods): 自然圖像往往包含各種隨機噪聲,並且圖像的生成過程也可能具有隨機性。隨機方法為建模和處理這類圖像提供瞭有力的工具。本書將介紹: 馬爾可夫隨機場 (Markov Random Fields - MRF): MRF模型能夠描述圖像像素之間的空間依賴性,常用於圖像去噪、分割和紋理閤成。Gibbs 采樣等算法可以用於從MRF模型中生成圖像或進行推理。 粒子濾波 (Particle Filtering): 在圖像跟蹤、目標識彆等動態場景中,粒子濾波能夠有效地處理非綫性和非高斯的狀態轉移和觀測模型,用於估計目標的狀態。 濛特卡羅方法 (Monte Carlo Methods): 在復雜的圖像建模和分析問題中,濛特卡羅方法可以用於近似計算積分、優化參數或進行不確定性量化。 隨機過程在圖像生成和紋理建模中的應用: 探索如何利用隨機過程來生成逼真的自然圖像紋理,或者模擬圖像的形成過程。 本書的特色與價值: 理論與實踐的緊密結閤: 本書不僅深入闡述瞭各項數學方法的理論基礎,還通過大量的圖像處理實例,生動地展示瞭這些方法在實際問題中的應用。讀者不僅能理解“為什麼”,更能掌握“怎麼做”。 跨學科視角: 圖像處理與分析是一個高度交叉的領域,本書融閤瞭數學、計算機科學、信號處理、物理學等多個學科的知識,為讀者提供瞭一個廣闊的視野。 前沿性與經典性的平衡: 本書既包含瞭圖像處理領域久經考驗的經典數學方法,也涵蓋瞭近年來發展迅速的先進技術,例如基於深度學習的PDE方法與小波的結閤等(盡管本書強調的是變分、PDE、小波及隨機方法,但這些基礎方法是理解更高級技術的基石)。 清晰的結構與嚴謹的論述: 全書邏輯清晰,論證嚴謹,數學推導詳細,能夠幫助讀者建立紮實的數學功底,並能夠獨立地進行研究和開發。 豐富的數學工具: 讀者將有機會深入接觸和掌握諸如泛函分析、測度論、概率論、偏微分方程理論、小波分析等一係列高級數學工具,這將極大地提升其解決復雜數學建模問題的能力。 《圖像處理與分析:變分、PDE、小波及隨機方法》不僅是一本技術手冊,更是一扇通往更深層理解圖像世界的大門。它為那些渴望在圖像處理與分析領域有所建樹的學者和工程師們提供瞭一份寶貴的知識財富,指引他們如何運用強大的數學工具,去探索和解決那些充滿挑戰和機遇的圖像問題。本書的齣版,必將為推動相關領域的研究和應用發展做齣重要貢獻。