Machine Vision Technology:Lecture9 Recognition | bag of features
- Introduction to object recognition
- Representation
- Learning
- Recognition
- Bag of Words models
- Basic representation
- Different learning and recognition algorithms
the different visual recognition tasks
- Classification:
Does this image contain a building? Is this an beach? [yes/no]
-
Image Search & Organizing photo collections
-
Detection:
Does this image contain a car? [where?]
Which object does this image contain? [where?]
Accurate localization (segmentation) :
Estimating object semantic & geometric attributes 估计对象的语义和几何属性
Applications of computer vision
- Computational photography 计算摄影:利用数字运算而非传统光学过程的数字图像拍摄。提高相机性能,减少成本和尺寸,生成3D图像、增强景深和选择性散焦等。
- Surveillance 监测监控
- Assistive driving 辅助驾驶
Categorization vs Single instance recognition
区分类别分类和单一实例识别。
The Goals of Visual Recognition
- Design algorithms that are capable to
- Classify images or videos
- Detect and localize objects
- Estimate semantic and geometrical attributes
- Classify human activities and events
The Challenges of Visual Recognition
-
viewpoint variation 视角变化
-
illumination 光照影响
-
scale 尺度影响
-
deformation 形变
-
occlusion 遮挡
- background clutter 背景杂波
- intra-class variation 类内变化
Basic issues
-
Representation
- How to represent an object category; which classification scheme?
-
Learning
- How to learn the classifier, given training data
-
Recognition
- How the classifier is to be used on novel data
Representation
- 提取区域:Building blocks: Sampling strategies 构建模块:抽样策略
提取点特征的算子称为兴趣算子或有利算子(Interest Operator),比较经典的有 Moravec 算子、Forstner 算子以及 Harris 算子等。
- 数据组织:Appearance only or location and appearance 仅外观或位置和外观
- Invariances for the challenges of Visual Recognition
To handle intra-class variability, it is convenient to describe an object categories using probabilistic models 为了处理类内的可变性,使用概率模型来描述对象类别是很方便的
Object models: Generative vs Discriminative vs hybrid 对象模型:生成型、判别型和混合型
- Object categorization: the statistical viewpoint 对象分类:统计学观点
Discriminative methods model posterior
Generative methods model likelihood and prior
- Discriminative models
- Generative models
Learning
Recognition
- Recognition task: classification, detection, etc. .
- Search strategy: Sliding Windows
- Simple
- Computational complexity ( x, y, S, θ, N of classes)
- BSW by Lampert et al 08
- Also, Alexe, et al 10
- Localization
- Objects are not boxes
- Prone to false positive
Bag-of-features models
Origin 1: Texture recognition
Origin 2: Bag-of-words models
Orderless document representation: frequencies of words from a dictionary 无序文档表示: 字典中单词的频率
Bag-of-features steps
- 1.Extract features : Regular grid or interest regions
- 2.Learn “visual vocabulary”
Example codebook:
- 3.Represent images by frequencies of “visual words”
Spatial pyramid representation 空间金字塔表示
Extension of a bag of features : Locally orderless representation at several levels of resolution
不同层级的向量合并起来构成更多维,即是两幅图像的level0相同,在更细的层次上level1可能不同。
HW
4.27 - 5.26