Computer Vision




Learning Intentions and Success Criteria

Learning Intentions

Success Criteria

In this tutorial I am going to learn :
  • how to ...
By the end of the tutorial I will be able to :
  • ...


The 5 stages of Computer Vision


1. Digitisation

The image source may be a digital camera, a video camera, a scanner or may be a satellite photograph. In the case of a video image, though digital video is becoming more common, it is likely that the analogue image will require to be converted, using a video digitiser, to a digital image.

The problems at this stage are connected with the quality of the image. This can be affected by electronic distortion, dust on the lens or poor lighting.

2. Signal Processing

The second stage is signal processing where the computer performs calculations on each pixel in the picture in order to enhance the quality of the image.

3. Edge and Region Detection


This involves the computer defining and locating objects and areas. The values of adjacent pixels are compared and so edges are detected. In both these stages there are a number of mathematical methods employed to analyse and interpret the pixels.

4. Object Recognition

Object recognition involves matching the pattern of edges and regions to stored patterns to identify individual objects in the whole image.

5. Image Understanding

Image understanding is putting together the individual objects, found at the fourth stage, into a comprehensible scene. Only then can our computer react to the scene.

What we require is a method of describing each object in our image and then to compare this description with a stored set of descriptions to obtain a match for our object.


Waltz Algorithm


This is a method of labelling edges according to whether they are concave edges (marked with a +), convex edges (marked with a -) or obscuring edges (marked with an arrow). The direction of the arrow on an obscuring edge indicates that the face in view is to the right while the obscured (or hidden) face is to the left.

You look at the image and try to categorise the vertex with the possible patterns in the bitmap.



We can inspect the labelled lines at each vertex and decide which of the above 18 possible patterns our vertex fits.
It may not be possible to categorise a vertex at first, but another vertex will supply information to categorise on a second or later repetition.

When successful matches are found, 'recognition' occurs - the computer identifies what it is looking at. This process is continued and the computer builds up a complete picture of the scene by matching each vertex and junction. By linking groups of vertices and using the rules (about particular type(s) of vertex associated with a particular shape) stored in its knowledge base, the system will be able to conclude that it has identified a cube, a cuboid or a pyramid, etc.

You probably realise that, as yet, vision systems have limited applications. The reason for this is the huge amount of information on objects and the processing required to identify them when we look at everyday scenes.

Imagine the complexity in simply extending the scope and trying to get a vision system to understand the scene in a room in your house. There would be a much greater number of objects (TV, video, settee, chairs, table, pictures on wall, plants, books, etc.). Thus a computer would have far more difficulty in picking out the various objects in the scene. If the vision system is to be used outside there will be further complications as objects like trees, bushes and animals will be very difficult to identify due to curved, irregular surfaces and the lack of hard edges.

Quick Quiz


Embed Discussion Board here

Chat Room


Embed Chat Room here

Video Conference


Embed Video Conference here

Additional Videos


Embed video here

Exam Questions


Embed Document here



Summary

  • Please make a list of essential key-points here


Files

  • upload files here



Links