Computer Vision:a Modern Approach 摘抄笔记——Chapter 9：Segmentation by Clustering

Chapter 9：Segmentation by Clustering

9.1   Human Vision: Grouping and Gestalt

A key feature of the human vision systemis that context affects how things are perceived.

A common experience of segmentation is the way that an image can resolve itself into a figure— typically, the significant, important object—and aground—the background on which the figure lies. 

The Gestalt school used the notion of a gestalt as central components in their ideas. Their work was characterized by attempts to write down a series of rules by which image elements would be associated together and interpreted as a group.

There are a variety of factors, some of which postdate the main Gestalt movement:

• Proximity: Tokens that are nearby tend to be grouped.
• Similarity: Similar tokens tend to be grouped together.
  
• Common fate: Tokens that have coherent motion tend to be grouped together.
  
• Common region: Tokens that lie inside the same closed region tend to be
  grouped together.
  
• Parallelism: Parallel curves or tokens tend to be grouped together.
  
• Closure: Tokens or curves that tend to lead to closed curves tend to be
  grouped together.
  
• Symmetry: Curves that lead to symmetric groups are grouped together.
  
• Continuity: Tokens that lead to continuous—as in joining up nicely, rather
  than in the formal sense—curves tend to be grouped.
  
• Familiar configuration: Tokens that, when grouped, lead to a familiar
  object tend to be grouped together.
  

    

               

但是，如何把上面这些rules用于形成算法还有难度，如无法把握何时选用哪条规则。

9.2   Important Applications

• 9.2.1 Background Subtraction

对于视频中背景移动改变等问题，一般用 a moving average 来估计背景像素的值。

• 9.2.2 Shot Boundary Detection
  

长video是由很多短镜头(shots) 组成的，每个镜头中大多物体是不变的，每个shot可以用一个关键帧表示。

A shot boundary detection algorithm must find frames in the video that are significantly different from the previous frame. 可以用distance表示，计算distance有几种方法，目前不太需要，此处略去，有需要可以翻看原书。

• 9.2.3 Interactive Segmentation
  
• 9.2.4 Forming Image Regions

9.3 IMAGE SEGMENTATION BY CLUSTERING PIXELS