具体描述
内容简介
数字视频是用数字手段提供全运动视频图象的高新技术，近十
余年来推助了多媒体，虚拟现实，视频通信，VCD等产业的飞速发
展；在即将来临的信息社会中，还将给计算机，通信，影象等产业
以巨大的推动。为帮助读者在未来破浪前进，这本及时问世的书首
次全面讲述了数字视频处理的原理以及面向各种应用的主要算法。
全书分为6个部分：数字视频表示，包括视频图象模型和空域一时
域采样；二维运动估计；三维运动估计；视频滤波；静图象压缩；
视频压缩。本书是在为研究主和高年级学生讲课星础上写成的，取
材全面系统，表述精练，插图丰富并有详尽的文献索引，对于所用
的数学原理，作者进行了仔细处理和精心安排，特别便于自学。
好的，这是一份关于其他主题的图书简介，专注于深度学习在自然语言处理中的应用，旨在提供详尽且富有洞察力的内容，且不提及您提到的“数字视频处理”： --- 深度自然语言理解：从Transformer到大型语言模型的范式革命 图书定价： 人民币 168.00 元 页数： 780 页 开本： 16 开 装帧： 胶装，配有精美插图和代码附录 内容提要 本书深入剖析了现代自然语言处理（NLP）领域的核心理论、关键算法以及前沿实践，尤其侧重于自注意力机制（Self-Attention）的出现如何彻底颠覆了我们对文本、语音和机器理解的传统认知。我们不再满足于基于循环（RNN）或卷积（CNN）的序列建模，而是全面转向基于Transformer 架构的深度神经网络，这不仅是技术的迭代，更是一场认知范式的革命。 全书内容组织结构严谨，从基础的词嵌入技术讲起，逐步过渡到复杂的预训练模型、指令微调（Instruction Tuning）以及当前主导人工智能领域的大型语言模型（LLMs）的构建、优化与部署。本书旨在为资深工程师、算法研究人员以及希望在AI前沿领域深耕的硕士和博士研究生提供一本全面、深入且极具实操指导价值的参考手册。 核心章节详解 第一部分：NLP基础与序列建模的演进（第1-4章） 本部分回顾了NLP从统计方法向深度学习过渡的历程。详细讲解了词向量（Word2Vec, GloVe）的局限性，并引入了上下文嵌入（Contextualized Embeddings）的概念，特别是ELMo和BERT的早期设计哲学。我们着重探讨了语言模型在信息抽取、命名实体识别（NER）和句法分析等传统任务中的表现，为理解后续Transformer的优越性打下坚实的基础。 第二部分：Transformer 架构的数学与工程精髓（第5-9章） 这是本书的核心技术篇章。我们对原始的 Attention Is All You Need 论文进行了彻底的解构与重建。 多头注意力机制（Multi-Head Attention）：详细推导了 Q、K、V 矩阵的线性代数运算，阐明了“多头”如何允许模型在不同的表示子空间中捕获信息。 位置编码（Positional Encoding）：深入分析了正弦/余弦编码与学习式位置嵌入的优劣，以及旋转位置嵌入（RoPE）在长序列处理中的优势。 Transformer 块的堆叠与优化：分析了残差连接（Residual Connections）、层归一化（Layer Normalization）的放置对训练稳定性的关键影响。 自回归与自编码模型：清晰区分了 GPT 族（仅解码器）和 BERT 族（编码器）的设计目标及其在生成与理解任务中的适用性。 第三部分：预训练与迁移学习的深度实践（第10-14章） 本部分聚焦于如何高效地利用海量无标签数据进行预训练，并实现跨任务的知识迁移。 掩码语言模型（MLM）与下一句预测（NSP）：详述 BERT 预训练策略的细节与挑战，特别是 NSP 在后续研究中被放弃的原因分析。 大规模语料的收集与清洗：探讨 Common Crawl、BooksCorpus 等数据集的特点，以及去重、质量过滤和Tokenization（如 BPE, WordPiece）的工程实践。 高效微调策略（Parameter-Efficient Fine-Tuning, PEFT）：重点介绍 LoRA (Low-Rank Adaptation) 和 Prompt Tuning 等技术，如何用极少的计算资源适配数十亿参数的模型到特定下游任务，这是当前工业界降低部署成本的关键技术。 多模态预训练的初步探索：简要介绍将文本与视觉信息（如 CLIP 的设计思想）结合的早期模型结构，为更广阔的AI应用铺路。 第四部分：大型语言模型（LLMs）的崛起与控制（第15-19章） 本部分深入当前研究的最前沿——万亿级参数模型的涌现能力、对齐问题与安全性考量。 规模法则（Scaling Laws）：分析计算资源、模型参数量与数据量对模型性能的幂律关系，指导资源分配决策。 指令遵循（Instruction Following）与人工反馈强化学习（RLHF）：详细阐述了 SFT（Supervised Fine-Tuning）后的关键步骤——如何通过人类偏好数据构建奖励模型（Reward Model），并使用 PPO 或 DPO 算法实现模型的“对齐”，使其行为更符合人类意图，而非仅仅是预测下一个词。 涌现能力（Emergent Abilities）的探讨：分析 CoT（Chain-of-Thought）提示、少样本学习（Few-Shot Learning）在大型模型中展现出的推理和规划能力，并提供构建有效复杂提示的框架。 模型安全与偏见减轻：系统讨论 LLMs 中存在的毒性、事实错误（幻觉）和固有偏见问题，并介绍后处理、输入过滤和对抗性训练等减轻策略。 第五部分：部署、推理与未来趋势（第20-22章） 收官部分侧重于如何将训练好的巨型模型投入实际应用，并展望技术发展方向。 模型量化与稀疏化：介绍 INT8, FP8 量化技术，以及权重剪枝（Pruning）在保持性能的同时，将模型部署到边缘设备或资源受限服务器上的可行性。 高效推理框架：分析 vLLM、FasterTransformer 等加速库的工作原理，重点讲解分页注意力（Paged Attention）机制如何解决 KV Cache 碎片化问题，实现高吞吐量。 检索增强生成（RAG）架构：讲解如何结合外部知识库（向量数据库）来增强 LLM 的事实准确性和知识时效性，这是当前企业级应用的首选范式。 本书特色 1. 理论深度与代码实践并重： 每项关键算法都附有清晰的数学推导，并配有基于 PyTorch 或 JAX 的伪代码/实战代码片段，方便读者复现和理解。 2. 聚焦前沿范式： 全书的重心明确放在 Transformer 及后续的 LLMs 及其对齐技术上，是理解当前工业界主流方向的必备读物。 3. 系统性与批判性思维： 不仅介绍“如何做”，更探讨“为什么这样做”，对当前热门技术（如 RLHF 的局限性）提出批判性分析。 本书是 NLP 领域研究人员和实践工程师的进阶指南，旨在帮助读者掌握从零开始构建、优化和部署下一代语言智能体的核心能力。
作者简介
目录信息
Contents
Preface
About the Author
About the Notation
REPRESENTATION OF DIGITAL VIDEO
1 BASICS OF VIDEO
1.1 Analog Video
1.1.1 Analog Video Signal
1.1.2 Analog Video Standards
1.1.3 Analog Video Equipment
1.2 Digital Video
1.2.1 Digital Video Signal
1.2.2 Digital Video Standards
1.2.3 Why Digital Video?
1.3 Digital Video Processing
2 TIME-VARYING IMAGE FORMATION MODELS
2.1 Three-Dimensional Motion Models
2.1.1 Rigid Motion in the Cartesian Coordinates
2.1.2 Rigid Motion in the Homogeneous Coordinates
2.1.3 Deformable Motion
2.2 Geometric Image Formation
2.2.1 Perspective Projection
2.2.2 Orthographic Projection
2.3 Photometric Image Formation
2.3.1 Lambertian Reflectance Model
2.3.2 Photometric Effects of 3-D Motion
2.4 Observation Noise
2.5 Exercises
3 SPATIO-TEMPORAL SAMPLING
3.1 Sampling for Analog and Digital Video
3.1.1 Sampling Structures for Analog Video
3.1.2 Sampling Structures for Digital Video
3.2 Two-Dimensional Rectangular Sampling
3.2.1 2-D Fourier Transform Relations
3.2.2 Spectrum of the Sampled Signal
3.3 Two-Dimensional Periodic Sampling
3.3.1 Sampling Geometry
3.3.2 2-D Fourier Transform Relations in Vector Form
3.3.3 Spectrum of the Sampled Signal
3.4 Sampling on 3-D Structures
3.4.1 Sampling on a Lattice
3.4.2 Fourier Transform on a Lattice
3.4.3 Spectrum of Signals Sampled on a Lattice
3.4.4 Other Sampling Structures
3.5 Reconstruction from Samples
3.5.1 Reconstruction from Rectangular Samples
3.5.2 Reconstruction from Samples on a Lattice
3.6 Exercises
4 SAMPLING STRUCTURE CONVERSION
4.1 Sampling Rate Change for l-D Signals
4.1.1 Interpolation of l-D Signals
4.1.2 Decimation of l-D Signals
4.1.3 Sampling Rate Change by a Rational Factor
4.2 Sampling Lattice Conversion
4.3 Exercises
5 TWO-DIMENSIONAL MOTION ESTIMATION
OPTICAL FLOW METHODS
5.1 2-D Motion vs. Apparent Motion
5.1.1 2-D Motion
5.1.2 Correspondence and Optical Flow
5.2 2-D Motion Estimation
5.2.1 The Occlusion Problem
5.2.2 The Aperture Problem
5.2.3 Two-Dimensional Motion Field Models
5.3 Methods Using the Optical Flow Equation
5.3.1 The Optical Flow Equation
5.3.2 Second-Order Differential Methods
5.3.3 Block Motion Model
5.3.4 Horn and Schunck Method
5.3.5 Estimation of the Gradients
5.3.6 Adaptive Methods
5.4 Examples
5.5 Exercises
6 BLOCK-BASED METHODS
6.1 Block-Motion Models
6.1.1 Translational Block Motion
6.1.2 Generalized/Deformable Block Motion
6.2 Phase-Correlation Method
6.2.1 The Phase-Correlation Function
6.2.2 Implementation Issues
6.3 Block-Matching Method
6.3.1 Matching Criteria
6.3.2 Search Procedures
6.4 Hierarchical Motion Estimation
6.5 Generalized Block-Motion Estimation
6.5.1 Postprocessing for Improved Motion Compensation
6.5.2 Deformable Block Matching
6.6 Examples
6.7 Exercises
7 PEL-RECURSIVE METHODS
7.1 Displaced Frame Difference
7.2 Gradient-Based Optimization
7.2.1 Steepest-Descent Method
7.2.2 Newton-Raphson Method
7.2.3 Local vs. Global Minima
7.3 Steepest-Descent-Based Algorithms
7.3.1 Netravali-Robbins Algorithm
7.3.2 Walker-Rao Algorithm
7.3.3 Extension to the Block Motion Model
7.4 Wiener-Estimation-Based Algorithms
7.5 Examples
7.6 Exercises
8 BAYESIAN METHODS
8.1 Optimization Methods
8.1.1 Simulated Annealing
8.1.2 Iterated Conditional Modes
8.1.3 Mean Field Annealing
8.1.4 Highest Confidence First
8.2 Basics of MAP Motion Estimation
8.2.1 The Likelihood Model
8.2.2 The Prior Model
8.3 MAP Motion Estimation Algorithms
8.3.1 Formulation with Discontinuity Model,
8.3.2 Estimation with Local Outlier Rejection
8.3.3 Estimation with Region Labeling
8.4 Examples
8.5 Exercises
III THREE-DIMENSIONAL MOTION ESTIMATION
AND SEGMENTATION
9 METHODS USING POINT CORRESPONDENCES
9.1 Modeling the Projected Displacement Field
9.1.1 Orthographic Displacement Field Model
9.1.2 Perspective Displacement Field Model
9.2 Methods Based on the Orthographic Model
9.2.1 Two-Step Iteration Method from Two Views
9.2.2 An Improved Iterative Method
9.3 Methods Based on the Perspective Model
9.3.1 The Epipolar Constraint and Essential Parameters
9.3.2 Estimation ofthe Essential Pararneters
9.3.3 Decomposition of the E-Matrix
9.3.4 Algorithm
9.4 The Case of 3-D Planar Surfaces
9.4.1 The Pure Parameters
9.4.2 Estimation ofthe Pure Parameters
9.4.3 Estimation ofthe Motion and Structure Parameters
9.5 Examples
9.5.1 Numerical Simulations
9.5.2 Experiments with Two Frames of Miss America
9.6 Exercises
10 OPTICAL FLOW AND DIRECT METHODS
10.1 Modeling the Projected Velocity Field
10.1.1 Orthographic Velocity Field Model
10.1.2 Perspective Velocity Field Model
10.1.3 Perspective Velocity vs. Displacement Models
10.2 Focus of Expansion
10.3 Algebraic Methods Using Optical Flow
10.3.1 Uniqueness of the Solution
10.3.2 Affine Flow
10.3.3 Quadratic Flow
10.3.4 Arbitrary Flow
10.4 Optimization Methods Using Optical Flow
10.5 Direct Methods
10.5.1 Extension ofOptical Flow-Based Methods
10.5.2 Tsai-Huang Method
10.6 Examples
10.6.1 Numerical Simulations
10.6.2 Experiments with Two Frames of Miss America
10.7 Exercises
11 MOTION SEGMENTATION
11.1 Direct Methods
11.1.1 Thresholding for Change Detection
11.1.2 An Algorithm Using Mapping Parameters
11.1.3 Estimation of Model Parameters
11.2 Optical Flow Segmentation
11.2.1 Modified Hough Transform Method
11.2.2 Segmentation for Layered Video Representation .
11.2.3 Bayesian Segmentation
11.3 Simultaneous Estimation and Segmentation
11.3.1 Motion Field Model
11.3.2 Problem Formulation
11.3.3 The Algorithm
11.3.4 Relationship to Other Algorithms
11.4 Examples
11.5 Exercises
12 STEREO AND MOTION TRACKING
12.1 Motion and Structure from Stereo
12.1.1 Still-Frame Stereo Imaging
12.1.2 3-D Feature Matching fbr Motion Estimation
12.1.3 Stereo-Motion Fusion
12.1.4 Extension to Multiple Motion
12.2 Motion Tracking
12.2.1 Basic Principles
12.2.2 2-D Motion Tracking
12.2.3 3-D Rigid Motion Ttacking
12.3 Examples
12.4 Exercises
13 MOTION COMPENSATED FILTERING
13.1 Spatio-Temporal Fourier Spectrum
13.1.1 Global Motion with Constant Velocity
13.1.2 Global Motion with Acceleration
13.2 Sub-Nyquist Spatio-Temporal Sampling
13.2.1 Sampling in the Temporal Direction Only
13.2.2 Sampling on a Spatio-Temporal Lattice
13.2.3 Critical Velocities
13.3 Filtering Along Motion TRajectories
13.3.1 Arbitrary Motion Trajectories
13.3.2 Global Motion with Constant Velocity
13.3.3 Accelerated Motion
13.4 Applications
13.4.1 Motion-Compensated Noise Filtering
13.4.2 Motion-Compensated Reconstruction Filtering
13.5 Exercises
14 NOISE FILTERING
14.1 Intraframe Filtering
14.1.1 LMMSE Filtering
14.1.2 Adaptive (Local) LMMSE Filtering
14.1.3 Directional Filtering
14.1.4 Median and Weighted Median Filtering
14.2 Motion-Adaptive Filtering
14.2.1 Direct Filtering
14.2.2 Motion-Detection Based Filtering
14.3 Motion-Compenaated Filtering
14.3.1 Spatio-Temporal Adaptive LMMSE Filtering
14.3.2 Adaptive Weighted Averaging Filter
14.4 Examples
14.5 Exercises
15 RESTORATION
15.1 Modeling
15.1.1 Shift-Invariant Spatial Blurring
15.1.2 Shift-Varying Spatial Blurring
15.2 Intraframe Shift-Invariant Restoration
15.2.1 Pseudo Inverse Filtering
15.2.2 Constrained Least Squares and Wiener Filtering
15.3 Intraframe Shift-Varying Restoration
15.3.1 Overview ofthe POCS Method
15.3.2 Restoration Using POCS
15.4 Multiframe Restoration
15.4.1 Cross-Correlated Multiframe Filter
15.4.2 Motion-Compensated Multiframe Filter
15.5 Examples
15.6 Exercises
16 STANDARDS CONVERSION
16.1 Down-Conversion
16.1.1 Down-Conversion with Anti-Alias Filtering
16.1.2 Down-Conversion without Anti-Alias Filtering
16.2 Practical Up-Conversion Methods
16.2.1 Intraframe Filtering
16.2.2 Motion-Adaptive Filtering
16.3 Motion-Compensated Up-Conversion
16.3.1 Basic Principles
16.3.2 Global-Motion-Compensated De-interlacing
16.4 Examples
16.5 Exercises
17 SUPERRESOLUTION
17.1 Modeling
17.1.1 Continuous-Discrete Model
17.1.2 Discrete-Discrete Model
17.1.3 Problem Interrelations
17.2 Interpolation-Restoration Methods
17.2.1 Intraframe Methods
17.2.2 Multiframe Methods
17.3 A Frequency Domain Method
17.4 A Unifying POCS Method
17.5 Examples
17.6 Exercises
V STILL IMAGE COMPRESSION
18 LOSSLESS COMPRESSION
18.1 Basics of Image Compression
18.1.1 Elements of an Image Compression System
18.1.2 Information Theoretic Concepts
18.2 Symbol Coding
18.2.1 Fixed-Length Coding
18.2.2 Huffman Coding
18.2.3 Arithmetic Coding
18.3 Lossless Compression Methods
18.3.1 Lossless Predictive Coding
18.3.2 Run-Length Coding of Bit-Planes
18.3.3 Ziv-Lempel Coding
18.4 Exercises
19 DPCM AND TRANSFORM CODING
19.1 Quantization
19.1.1 Nonuniform Quantization
19.1.2 Uniform Quantization
19.2 Differential Pulse Code Modulation
19.2.1 Optimal Prediction
19.2.2 Quantization of the Prediction Error
19.2.3 Adaptive Quantization
19.2.4 Delta Modulation
19.3 Transform Coding
19.3.1 Discrete Cosine Transform
19.3.2 Quantization/Bit Allocation
19.3.3 Coding
19.3.4 Blocking Artifacts in Transform Coding
19.4 Exercises
20 STILL IMAGE COMPRESSION STANDARDS
20.1 Bilevel Image Compression Standards
20.1.1 One-Dimensional RLC
20.1.2 Two-Dimensional RLC
20.1.3 The JBIG Standard
20.2 The JPEG Standard
20.2.1 Baseline Algorithm
20.2.2 JPEG Progressive
20.2.3 JPEG Lossless
20.2.4 JPEG Hierarchical
20.2.5 ImplementationsofJPEG
20.3 Exercises
21 VECTOR QUANTIZATION, SUBBAND CODING
AND OTHER METHODS
21.1 Vector Quantization
21.1.1 Structure of a Vector Quantizer
21.1.2 VQ Codebook Design
21.1.3 Practical VQ Implementations
21.2 Fractal Compression
21.3 Subband Coding
21.3.1 Subband Decomposition
21.3.2 Coding of the Subbands
21.3.3 Relationship to Transform Coding
21.3.4 Relationship to Wavelet Transform Coding
21.4 Second-Generation Coding Methods
21.5 Exercises
VI VIDEO COMPRESSION
22 INTERFRAME COMPRESSION METHODS
22.1 Three-Dimensional Waveform Coding
22.1.1 3-D Transform Coding
22.1.2 3-D Subband Coding
22.2 Motion-Compensated Waveform Coding
22.2.1 MC Transform Coding
22.2.2 MC Vector Quantization
22.2.3 MC Subband Coding
22.3 Model-Based Coding
22.3.1 Object-Based Coding
22.3.2 Knowledge-Based and Semantic Coding
22.4 Exerclses
23 VIDEO COMPRESSION STANDARDS
23.1 The H.261 Standard
23.1.1 Input Image Formats
23.1.2 Video Multiplex
23.1.3 Video Compression Algorithm
23.2 The MPEG-l Standard
23.2.1 Features
23.2.2 Input Video Format
23.2.3 Data Structure and Compression Modes
23.2.4 Intraframe Compression Mode
23.2.5 Interframe Compression Modes
23.2.6 MPEG-l Encoder and Decoder
23.3 The MPEG-2 Standard
23.3.1 MPEG-2 Macroblocks
23.3.2 Coding Interlaced Video
23.3.3 Scalable Extensions
23.3.4 Other Improvements
23.3.5 Overview of Profiles and Levels
23.4 Software and Hardware Implementations
24 MODEL-BASED CODING
24.1 General Object-Based Methods
24.1.1 2-D/3-D Rigid Objects with 3-DMotion
24.1.2 2-D Flexible Objects with 2-D Motion
24.1.3 Affine Transformations with TRiangular Meshes
24.2 Knowledge-Based and Semantic Methods
24.2.1 General Principles
24.2.2 MBASIC Algorithm
24.2.3 Estimation Using a Flexible Wireframe Model
24.3 Examples
25 DIGITAL VIDEO SYSTEMS
25.1 Videoconferencing
25.2 Interactive Video and Multimedia
25.3 Digital Television
25.3.1 Oigital Studio Standards
25.3.2 Hybrid Advanced TV Systems
25.3.3 All-Oigital TV
25.4 Low-Bitrate Video and Videophone
25.4.1 The ITU Recommendation H.263
25.4.2 The ISO MPEG-4 Requirements
APPENDICES
A MARKOV AND GIBBS RANDOM FIELDS
A.l Definitions
A.l.l Markov Random Fields
A.1.2 Gibbs Random Fields
A.2 Equivalence of MRF and GRF
A.3Local Conditional Probabilities
B BASICS OF SEGMENTATION
B.l Thresholding
B.I.l Finding the Optimum Threshold(s)
B.2 Clustering
B.3 Bayesian Methods
B.3.1 The MAP Method
B.3.2 The Adaptive MAP Method
B.3.3 Vector Field Segmentation
C KALMAN FILTEMNG
C.l Linear State-Space Model
C.2 Extended Kalman Filtering
· · · · · · (收起)
Preface
About the Author
About the Notation
REPRESENTATION OF DIGITAL VIDEO
1 BASICS OF VIDEO
1.1 Analog Video
1.1.1 Analog Video Signal
1.1.2 Analog Video Standards
1.1.3 Analog Video Equipment
1.2 Digital Video
1.2.1 Digital Video Signal
1.2.2 Digital Video Standards
1.2.3 Why Digital Video?
1.3 Digital Video Processing
2 TIME-VARYING IMAGE FORMATION MODELS
2.1 Three-Dimensional Motion Models
2.1.1 Rigid Motion in the Cartesian Coordinates
2.1.2 Rigid Motion in the Homogeneous Coordinates
2.1.3 Deformable Motion
2.2 Geometric Image Formation
2.2.1 Perspective Projection
2.2.2 Orthographic Projection
2.3 Photometric Image Formation
2.3.1 Lambertian Reflectance Model
2.3.2 Photometric Effects of 3-D Motion
2.4 Observation Noise
2.5 Exercises
3 SPATIO-TEMPORAL SAMPLING
3.1 Sampling for Analog and Digital Video
3.1.1 Sampling Structures for Analog Video
3.1.2 Sampling Structures for Digital Video
3.2 Two-Dimensional Rectangular Sampling
3.2.1 2-D Fourier Transform Relations
3.2.2 Spectrum of the Sampled Signal
3.3 Two-Dimensional Periodic Sampling
3.3.1 Sampling Geometry
3.3.2 2-D Fourier Transform Relations in Vector Form
3.3.3 Spectrum of the Sampled Signal
3.4 Sampling on 3-D Structures
3.4.1 Sampling on a Lattice
3.4.2 Fourier Transform on a Lattice
3.4.3 Spectrum of Signals Sampled on a Lattice
3.4.4 Other Sampling Structures
3.5 Reconstruction from Samples
3.5.1 Reconstruction from Rectangular Samples
3.5.2 Reconstruction from Samples on a Lattice
3.6 Exercises
4 SAMPLING STRUCTURE CONVERSION
4.1 Sampling Rate Change for l-D Signals
4.1.1 Interpolation of l-D Signals
4.1.2 Decimation of l-D Signals
4.1.3 Sampling Rate Change by a Rational Factor
4.2 Sampling Lattice Conversion
4.3 Exercises
5 TWO-DIMENSIONAL MOTION ESTIMATION
OPTICAL FLOW METHODS
5.1 2-D Motion vs. Apparent Motion
5.1.1 2-D Motion
5.1.2 Correspondence and Optical Flow
5.2 2-D Motion Estimation
5.2.1 The Occlusion Problem
5.2.2 The Aperture Problem
5.2.3 Two-Dimensional Motion Field Models
5.3 Methods Using the Optical Flow Equation
5.3.1 The Optical Flow Equation
5.3.2 Second-Order Differential Methods
5.3.3 Block Motion Model
5.3.4 Horn and Schunck Method
5.3.5 Estimation of the Gradients
5.3.6 Adaptive Methods
5.4 Examples
5.5 Exercises
6 BLOCK-BASED METHODS
6.1 Block-Motion Models
6.1.1 Translational Block Motion
6.1.2 Generalized/Deformable Block Motion
6.2 Phase-Correlation Method
6.2.1 The Phase-Correlation Function
6.2.2 Implementation Issues
6.3 Block-Matching Method
6.3.1 Matching Criteria
6.3.2 Search Procedures
6.4 Hierarchical Motion Estimation
6.5 Generalized Block-Motion Estimation
6.5.1 Postprocessing for Improved Motion Compensation
6.5.2 Deformable Block Matching
6.6 Examples
6.7 Exercises
7 PEL-RECURSIVE METHODS
7.1 Displaced Frame Difference
7.2 Gradient-Based Optimization
7.2.1 Steepest-Descent Method
7.2.2 Newton-Raphson Method
7.2.3 Local vs. Global Minima
7.3 Steepest-Descent-Based Algorithms
7.3.1 Netravali-Robbins Algorithm
7.3.2 Walker-Rao Algorithm
7.3.3 Extension to the Block Motion Model
7.4 Wiener-Estimation-Based Algorithms
7.5 Examples
7.6 Exercises
8 BAYESIAN METHODS
8.1 Optimization Methods
8.1.1 Simulated Annealing
8.1.2 Iterated Conditional Modes
8.1.3 Mean Field Annealing
8.1.4 Highest Confidence First
8.2 Basics of MAP Motion Estimation
8.2.1 The Likelihood Model
8.2.2 The Prior Model
8.3 MAP Motion Estimation Algorithms
8.3.1 Formulation with Discontinuity Model,
8.3.2 Estimation with Local Outlier Rejection
8.3.3 Estimation with Region Labeling
8.4 Examples
8.5 Exercises
III THREE-DIMENSIONAL MOTION ESTIMATION
AND SEGMENTATION
9 METHODS USING POINT CORRESPONDENCES
9.1 Modeling the Projected Displacement Field
9.1.1 Orthographic Displacement Field Model
9.1.2 Perspective Displacement Field Model
9.2 Methods Based on the Orthographic Model
9.2.1 Two-Step Iteration Method from Two Views
9.2.2 An Improved Iterative Method
9.3 Methods Based on the Perspective Model
9.3.1 The Epipolar Constraint and Essential Parameters
9.3.2 Estimation ofthe Essential Pararneters
9.3.3 Decomposition of the E-Matrix
9.3.4 Algorithm
9.4 The Case of 3-D Planar Surfaces
9.4.1 The Pure Parameters
9.4.2 Estimation ofthe Pure Parameters
9.4.3 Estimation ofthe Motion and Structure Parameters
9.5 Examples
9.5.1 Numerical Simulations
9.5.2 Experiments with Two Frames of Miss America
9.6 Exercises
10 OPTICAL FLOW AND DIRECT METHODS
10.1 Modeling the Projected Velocity Field
10.1.1 Orthographic Velocity Field Model
10.1.2 Perspective Velocity Field Model
10.1.3 Perspective Velocity vs. Displacement Models
10.2 Focus of Expansion
10.3 Algebraic Methods Using Optical Flow
10.3.1 Uniqueness of the Solution
10.3.2 Affine Flow
10.3.3 Quadratic Flow
10.3.4 Arbitrary Flow
10.4 Optimization Methods Using Optical Flow
10.5 Direct Methods
10.5.1 Extension ofOptical Flow-Based Methods
10.5.2 Tsai-Huang Method
10.6 Examples
10.6.1 Numerical Simulations
10.6.2 Experiments with Two Frames of Miss America
10.7 Exercises
11 MOTION SEGMENTATION
11.1 Direct Methods
11.1.1 Thresholding for Change Detection
11.1.2 An Algorithm Using Mapping Parameters
11.1.3 Estimation of Model Parameters
11.2 Optical Flow Segmentation
11.2.1 Modified Hough Transform Method
11.2.2 Segmentation for Layered Video Representation .
11.2.3 Bayesian Segmentation
11.3 Simultaneous Estimation and Segmentation
11.3.1 Motion Field Model
11.3.2 Problem Formulation
11.3.3 The Algorithm
11.3.4 Relationship to Other Algorithms
11.4 Examples
11.5 Exercises
12 STEREO AND MOTION TRACKING
12.1 Motion and Structure from Stereo
12.1.1 Still-Frame Stereo Imaging
12.1.2 3-D Feature Matching fbr Motion Estimation
12.1.3 Stereo-Motion Fusion
12.1.4 Extension to Multiple Motion
12.2 Motion Tracking
12.2.1 Basic Principles
12.2.2 2-D Motion Tracking
12.2.3 3-D Rigid Motion Ttacking
12.3 Examples
12.4 Exercises
13 MOTION COMPENSATED FILTERING
13.1 Spatio-Temporal Fourier Spectrum
13.1.1 Global Motion with Constant Velocity
13.1.2 Global Motion with Acceleration
13.2 Sub-Nyquist Spatio-Temporal Sampling
13.2.1 Sampling in the Temporal Direction Only
13.2.2 Sampling on a Spatio-Temporal Lattice
13.2.3 Critical Velocities
13.3 Filtering Along Motion TRajectories
13.3.1 Arbitrary Motion Trajectories
13.3.2 Global Motion with Constant Velocity
13.3.3 Accelerated Motion
13.4 Applications
13.4.1 Motion-Compensated Noise Filtering
13.4.2 Motion-Compensated Reconstruction Filtering
13.5 Exercises
14 NOISE FILTERING
14.1 Intraframe Filtering
14.1.1 LMMSE Filtering
14.1.2 Adaptive (Local) LMMSE Filtering
14.1.3 Directional Filtering
14.1.4 Median and Weighted Median Filtering
14.2 Motion-Adaptive Filtering
14.2.1 Direct Filtering
14.2.2 Motion-Detection Based Filtering
14.3 Motion-Compenaated Filtering
14.3.1 Spatio-Temporal Adaptive LMMSE Filtering
14.3.2 Adaptive Weighted Averaging Filter
14.4 Examples
14.5 Exercises
15 RESTORATION
15.1 Modeling
15.1.1 Shift-Invariant Spatial Blurring
15.1.2 Shift-Varying Spatial Blurring
15.2 Intraframe Shift-Invariant Restoration
15.2.1 Pseudo Inverse Filtering
15.2.2 Constrained Least Squares and Wiener Filtering
15.3 Intraframe Shift-Varying Restoration
15.3.1 Overview ofthe POCS Method
15.3.2 Restoration Using POCS
15.4 Multiframe Restoration
15.4.1 Cross-Correlated Multiframe Filter
15.4.2 Motion-Compensated Multiframe Filter
15.5 Examples
15.6 Exercises
16 STANDARDS CONVERSION
16.1 Down-Conversion
16.1.1 Down-Conversion with Anti-Alias Filtering
16.1.2 Down-Conversion without Anti-Alias Filtering
16.2 Practical Up-Conversion Methods
16.2.1 Intraframe Filtering
16.2.2 Motion-Adaptive Filtering
16.3 Motion-Compensated Up-Conversion
16.3.1 Basic Principles
16.3.2 Global-Motion-Compensated De-interlacing
16.4 Examples
16.5 Exercises
17 SUPERRESOLUTION
17.1 Modeling
17.1.1 Continuous-Discrete Model
17.1.2 Discrete-Discrete Model
17.1.3 Problem Interrelations
17.2 Interpolation-Restoration Methods
17.2.1 Intraframe Methods
17.2.2 Multiframe Methods
17.3 A Frequency Domain Method
17.4 A Unifying POCS Method
17.5 Examples
17.6 Exercises
V STILL IMAGE COMPRESSION
18 LOSSLESS COMPRESSION
18.1 Basics of Image Compression
18.1.1 Elements of an Image Compression System
18.1.2 Information Theoretic Concepts
18.2 Symbol Coding
18.2.1 Fixed-Length Coding
18.2.2 Huffman Coding
18.2.3 Arithmetic Coding
18.3 Lossless Compression Methods
18.3.1 Lossless Predictive Coding
18.3.2 Run-Length Coding of Bit-Planes
18.3.3 Ziv-Lempel Coding
18.4 Exercises
19 DPCM AND TRANSFORM CODING
19.1 Quantization
19.1.1 Nonuniform Quantization
19.1.2 Uniform Quantization
19.2 Differential Pulse Code Modulation
19.2.1 Optimal Prediction
19.2.2 Quantization of the Prediction Error
19.2.3 Adaptive Quantization
19.2.4 Delta Modulation
19.3 Transform Coding
19.3.1 Discrete Cosine Transform
19.3.2 Quantization/Bit Allocation
19.3.3 Coding
19.3.4 Blocking Artifacts in Transform Coding
19.4 Exercises
20 STILL IMAGE COMPRESSION STANDARDS
20.1 Bilevel Image Compression Standards
20.1.1 One-Dimensional RLC
20.1.2 Two-Dimensional RLC
20.1.3 The JBIG Standard
20.2 The JPEG Standard
20.2.1 Baseline Algorithm
20.2.2 JPEG Progressive
20.2.3 JPEG Lossless
20.2.4 JPEG Hierarchical
20.2.5 ImplementationsofJPEG
20.3 Exercises
21 VECTOR QUANTIZATION, SUBBAND CODING
AND OTHER METHODS
21.1 Vector Quantization
21.1.1 Structure of a Vector Quantizer
21.1.2 VQ Codebook Design
21.1.3 Practical VQ Implementations
21.2 Fractal Compression
21.3 Subband Coding
21.3.1 Subband Decomposition
21.3.2 Coding of the Subbands
21.3.3 Relationship to Transform Coding
21.3.4 Relationship to Wavelet Transform Coding
21.4 Second-Generation Coding Methods
21.5 Exercises
VI VIDEO COMPRESSION
22 INTERFRAME COMPRESSION METHODS
22.1 Three-Dimensional Waveform Coding
22.1.1 3-D Transform Coding
22.1.2 3-D Subband Coding
22.2 Motion-Compensated Waveform Coding
22.2.1 MC Transform Coding
22.2.2 MC Vector Quantization
22.2.3 MC Subband Coding
22.3 Model-Based Coding
22.3.1 Object-Based Coding
22.3.2 Knowledge-Based and Semantic Coding
22.4 Exerclses
23 VIDEO COMPRESSION STANDARDS
23.1 The H.261 Standard
23.1.1 Input Image Formats
23.1.2 Video Multiplex
23.1.3 Video Compression Algorithm
23.2 The MPEG-l Standard
23.2.1 Features
23.2.2 Input Video Format
23.2.3 Data Structure and Compression Modes
23.2.4 Intraframe Compression Mode
23.2.5 Interframe Compression Modes
23.2.6 MPEG-l Encoder and Decoder
23.3 The MPEG-2 Standard
23.3.1 MPEG-2 Macroblocks
23.3.2 Coding Interlaced Video
23.3.3 Scalable Extensions
23.3.4 Other Improvements
23.3.5 Overview of Profiles and Levels
23.4 Software and Hardware Implementations
24 MODEL-BASED CODING
24.1 General Object-Based Methods
24.1.1 2-D/3-D Rigid Objects with 3-DMotion
24.1.2 2-D Flexible Objects with 2-D Motion
24.1.3 Affine Transformations with TRiangular Meshes
24.2 Knowledge-Based and Semantic Methods
24.2.1 General Principles
24.2.2 MBASIC Algorithm
24.2.3 Estimation Using a Flexible Wireframe Model
24.3 Examples
25 DIGITAL VIDEO SYSTEMS
25.1 Videoconferencing
25.2 Interactive Video and Multimedia
25.3 Digital Television
25.3.1 Oigital Studio Standards
25.3.2 Hybrid Advanced TV Systems
25.3.3 All-Oigital TV
25.4 Low-Bitrate Video and Videophone
25.4.1 The ITU Recommendation H.263
25.4.2 The ISO MPEG-4 Requirements
APPENDICES
A MARKOV AND GIBBS RANDOM FIELDS
A.l Definitions
A.l.l Markov Random Fields
A.1.2 Gibbs Random Fields
A.2 Equivalence of MRF and GRF
A.3Local Conditional Probabilities
B BASICS OF SEGMENTATION
B.l Thresholding
B.I.l Finding the Optimum Threshold(s)
B.2 Clustering
B.3 Bayesian Methods
B.3.1 The MAP Method
B.3.2 The Adaptive MAP Method
B.3.3 Vector Field Segmentation
C KALMAN FILTEMNG
C.l Linear State-Space Model
C.2 Extended Kalman Filtering
· · · · · · (收起)
读后感
评分
评分
评分
评分
评分
用户评价
评分
这本书给我带来的最大收获，在于它构建了一个“批判性思维”的框架。它不仅仅是教你如何使用现有的工具或算法，更重要的是，它引导你去思考这些技术本身的局限性和未来的发展方向。在讨论了现有的视频编码标准及其面临的挑战后，书中专门开辟了一章，展望了诸如基于神经网络的视频压缩等新兴领域的研究热点。这种前瞻性的内容设置，对于身处技术快速迭代行业的专业人士来说，价值不可估量。它让我意识到，视频处理技术并非一成不变的铁律，而是不断在追求更低码率、更高质量和更低延迟之间寻求平衡的动态过程。更细致地讲，书中对码率控制策略的探讨，就充分体现了这种辩证思维——它清晰地指出了固定码率、可变码率以及CBR/VBR之间的取舍，并分析了在不同网络带宽限制下，每种策略的优缺点。这种不带偏见的深入剖析，使得读者能够基于充分的理解做出更明智的技术决策，而不是盲目跟从标准。这本书，无疑是拓宽视野、深化理解的绝佳选择。
评分这本书的排版和插图设计，也绝对值得称赞。在处理如此高密度的技术信息时，清晰的视觉呈现是至关重要的。我发现，作者对图表的运用达到了炉火纯青的地步。与那些密密麻麻的文字段落相比，书中大量的流程图、时域-空域关系图以及变换域的示意图，极大地减轻了阅读负担。例如，在解释离散余弦变换（DCT）时，配上的那组不同频率分量的可视化图谱，简直是神来之笔。它直观地展示了高频系数代表了哪些细节信息，以及在压缩过程中丢弃高频系数对人眼感知的具体影响。这种“所见即所得”的教学方式，比单纯的文字描述效率高出太多。此外，书中对关键术语的定义和脚注处理得非常专业，使得读者在阅读过程中可以随时查阅补充信息，而不会被冗长的解释打断思维的连贯性。总而言之，这是一本让人愿意反复翻阅、值得珍藏的工具书，它的视觉呈现提升了学习体验的质量。
评分我对这本书的评价，必须着重提到它在理论深度与工程实践之间的完美平衡。很多学术著作过于偏重数学推导，读起来晦涩难懂，让人觉得脱离实际；而市面上很多“速成”类的书籍又过于肤浅，只停留在软件调参层面，无法触及核心原理。这本书则像是一座坚固的桥梁，将两者优雅地连接起来。举个例子，当书中谈及去块效应（Deblocking Filter）时，它不仅给出了滤波器的数学模型，还结合实际视频序列的特点，解释了为什么在某些低码率场景下，这种后处理技术至关重要，以及如何根据不同的场景调整滤波器的强度。这种“理论+应用场景分析”的写法，使得书中的每一个公式和每一个参数都有了具体的“落脚点”。我尝试用书中提供的思路去分析一些开源项目中的视频解码器代码，惊讶地发现，那些曾经让我望而生畏的代码片段，现在竟能通过书中的理论框架被一一解读。这说明作者不仅是一位理论家，更是一位深谙工程实践的专家，能够准确地捕捉到理论在实际工程中遇到的痛点，并提供解决方案。
评分这本《数字视频处理》的书籍，坦率地说，我拿到手里的时候，心里是有些忐忑的。毕竟“数字视频处理”这个领域听起来就充满了复杂的数学公式和晦涩难懂的专业术语，我担心自己一个初学者根本啃不动。然而，真正翻开它之后，我的顾虑很快就被打消了。作者在开篇并没有直接抛出那些硬核的技术细节，而是用一种非常平易近人的方式，勾勒出了数字视频从采集到显示的整个流程图景。他巧妙地将那些复杂的理论知识“包装”成了更容易理解的实际应用案例。比如，在讲解采样和量化的时候，作者没有仅仅停留在理论公式上，而是举例说明了为什么高清电视和标清电视在画面细节上有天壤之别，以及这背后涉及到的比特率和颜色空间转换的实际意义。这种由宏观到微观的叙事结构，极大地降低了学习曲线。我尤其欣赏书中对“时间域”和“空间域”概念的阐述，讲解得深入浅出，仿佛是身边一位经验丰富的工程师在耐心地为你拆解难题。对于那些希望从零开始构建自己视频处理知识体系的读者来说，这本书无疑是一个非常扎实且引人入胜的起点，它没有让人在专业术语的迷雾中迷失方向，反而像一盏明灯，清晰地指引着前方的道路。
评分说实话，我过去也尝试过几本号称是“入门”的视频处理教材，但读完之后感觉知识点是零散的，就像一堆未经整理的零件，虽然工具都在，但不知道该如何组装成一个能运转的系统。但这本书的结构设计，简直是教科书级别的典范。它不是简单地罗列算法，而是构建了一个完整的知识体系的骨架。它清晰地划分了预处理、压缩、传输和播放等几个核心模块，并且在每个模块内部，都做到了逻辑上的严丝合缝。最让我印象深刻的是它在讲解视频压缩标准（比如H.264或HEVC的关键部分）时，那种层层递进的逻辑推导。作者并没有直接给出最终的编码器结构，而是先解释了为什么需要预测编码，然后引出运动估计和补偿的必要性，最后才详述了残差编码和熵编码的作用。这种“发现问题——分析问题——解决问题”的叙事方式，让我真正理解了这些复杂技术背后的“为什么”，而非仅仅是“是什么”。读完这一部分，我感觉自己不仅是学会了几个概念，而是对整个视频流的生命周期有了一个全局的、结构化的认识，这对后续深入研究更前沿的算法打下了坚实的基础。
评分 评分 评分 评分 评分相关图书
本站所有内容均为互联网搜索引擎提供的公开搜索信息，本站不存储任何数据与内容，任何内容与数据均与本站无关，如有需要请联系相关搜索引擎包括但不限于百度，google,bing,sogou 等
© 2026 qciss.net All Rights Reserved. 小哈图书下载中心 版权所有