Notes on Computer Vision A Modern Approach 2E

A: What do you want from me?

What should I know to consider myself expert in CV?

A: How an object is separated from its background?

An object is separated from its background in an image by an occluding contour.

A: What would you want from Chapter 1?

Chapter 1 is about cameras and their parameters. I don't like to learn much about these at the moment.

A: What would you want from Chapter 2?

Chapter 2 is about light and shading. It's also about cameras and looks like to build on chapter 1. I don't like to learn about it at the moment.

A: What are linear filters?

Linear filters are convolution based filters, like Gaussian. The chapter tells the theoretical basis of these filters.

A: What are local features?

Local features are (mostly) gradient based features which are computed using colvolving a derivation filter. SIFT, HoG are examples of such features. We can also have orientation histograms for images.

A: How may I capture texture information?

The unit of information in textures is called texton. They can be produced by the response of certain filters. There are many different filters and their responses change very much. There are also techniques for discovering textons. For holes and image generation, texton synthesis is also possible.

A: How HoG works?

It's a variant of SIFT, but normalization occurs locally so that lower contrast are caught better in HoG.

A: How textons are discovered?

Textons can be discovered using predefined filters. Response to these filters can be used to identify textons.

A: What can I learn about document recognition?

Mid Level Vision may be important. Classification chapters. Registration. Curve and line fitting.

A: How curve fitting works?

The methods for fitting curves is complex and does not result in robust solutions. It might be preferable to represent the curve as a set of line segments to overcome this.

A: How line fitting works?

Lines are fit using least squares method. In the first version, only the vertical distances to lines are considered. In the second version, total distance (perpendicular distance) is considered.

A: What are the approaches in segmentation?

I only took a look at gestalt ideas.

A: Which classification approaches are discussed?

There is not much info about this. I'm fed up with the book.