Note《Anchored Neighborhood Regression for Fast Example-Based Super-Resolution》
来源:互联网 发布:淘宝买号被找回 编辑:程序博客网 时间:2024/06/06 08:47
Note:
Anchored NeighborhoodRegression for Fast Example-Based Super-Resolution
1. Abstract
a) Propose fast super-resolutionmethods while making no compromise on quality
i. Support the use of sparselearned dictionaries in combination with neighbor embedding methods.
1). Dictionary atomsßEuclidean distance
ii. Use global collaborative coding
iii. Propose the anchoredneighborhood regression
1). Anchor the neighborhoodembedding
2). Precompute the correspondingembedding matrix
2. Introduction
a) Defination of super-resolution
b) Three subclasses
i. Interpilation methods
ii. Multi-frame methods
iii. Learning-based methods
1). Gradient Profile Prior
2). Dictionary- or example- learning methods
a) Subdivided into patches
b) Form a Markov Random Field(MRF)
c) Search for nearest neighbors
d) HR is retrieved
e) MRF can be solved
3). Downside
a) High computational complexity
b) Overcome:
i. Neighbor embedding
ii. Sparse encoding approaches
4). Proposed example-basedsuper-resolution
a). Low computational time
b). Qualitative performance
c) Organization
i. Section 2:neighbor embedding& sparse coding
ii. Section 3: proposed methods
iii. Section 4:experimental results
iv. Section 5:conclusions
3. Dictionary-basedSuper-Resolution
a) Neighbor embedding approaches
i. Low-dimensional nonlinearmanifolds
ii. Locally linear embedding(LLE)
1. Search for a set of K nearestneighbors
2. Compute K appropriate weights
3. Create HR patchs
4. Create result HR image
iii. Nonnegative neighbor embeddingapproaches
b) Sparse coding approaches
i. Effects: a learned compactdictionary
ii.
iii. sparsedictionaries:
iv. several modifications:
1). different training approaches
2). pseudoinverse(伪逆法)
3). PCA
4). Orthogonal matching pursuit
4. Proposed Methods
a) Global regression :special caseof ANR
i.
ii.
iii.
iv.
v.
vi.
b) Anchored neighborhoodregression
5. Experiments
a) Conditions
i. Features
1). Luminance component
2). Basic feature: the patch
3). First and second order derivative
ii. Embeddings
iii. Dictionaries
1). The larger the dictionary thebetter the performance
2). “internal”dictionary,”external” dictionary
3). Randomly sampled dictionaries,learned dictionaries
iv. Neighborhoods
b) Performance
i. Quality
ii. Running times
6. Conclusions
a) Propose a new example-basedmethod for super-resolution called Anchored Neighbor Regression
b) Propose an extreme variantcalled Global Regression
c) Most of these can reach asimilar top performance based on using the appropriate neighborhood size anddictionary
- Note《Anchored Neighborhood Regression for Fast Example-Based Super-Resolution》
- Seven ways to improve example-based single image super resolution笔记
- 图像超分辨LapSRN:Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution论文笔记
- Fast and Accurate Image Upscaling with Super-Resolution Forests
- Fast and Accurate Image Upscaling with Super-Resolution Forests
- super-resolution
- 《Vehicle Detection in High-Resolution Aerial Images Based on Fast Sparse Representation ...》论文笔记
- Learning a Deep Convolutional Network for Image Super-Resolution(泛读)
- Deeply-Recursive Convolutional Network for Image Super-Resolution 笔记
- Learning a Deep Convolutional Network for Image Super-Resolution(泛读)
- Enhanced Deep Residual Networks for Single Image Super-Resolution
- Deeply-Recursive Convolutional Network for Image Super-Resolution
- 【超分辨率】Deeply-Recursive Convolutional Network for Image Super-Resolution
- Enhanced Deep Residual Networks for Single Image Super-Resolution
- Paper Review-Fast and Robust Multiframe Super Resolution-#1-"Brain" hang out
- “Nonparametric Blind Super-Resolution”
- [gpu pro]Fast, Stencil-Based Multiresolution Splatting for Indirect Illumination
- Fast R-CNN: Fast Region-based Convolutional Networks for object detection
- spring学习笔记(10)@AspectJ研磨分析[1]入门、注解基本介绍
- Java以逗号分隔而组成的两个字符串,获取相同元素的字符串
- 【3】Nginx服务器基本配置简述
- CCD及CMOS尺寸计算
- 《网站设计 网页配色》笔记
- Note《Anchored Neighborhood Regression for Fast Example-Based Super-Resolution》
- 如何解决多线程程序中的死锁问题
- Android硬件监听
- H5 - HTML5新增标签
- 下表操作符
- hdoj--4505--小Q系列故事――电梯里的爱情(水题)
- Java实现的归并排序
- TCP传输控制协议总结
- Java核心知识点-类加载器