PART 2 Intermediate-Level Vision

Intermediate-level image analysis is concerned with obtaining abstract information about images, starting with the images themselves. At this stage we are no longer interested in converting one image into another, as in the subject of image processing. In fact, intermediate-level processing was already being developed in parts of Chapter 7. However, what is special about Part 2 is that intermediate-level analysis is investigated for its own sake, in particular with the aid of transform methods that have been designed systematically for the purpose.

The chapters in Part 2 largely result in abstract information on the positions and orientations of various image features: they do not aim to provide real-world data, a function that is left to Parts 3 and 4. (Thus, Part 2 may indicate that a circle exists in one part of an image; it is left to later parts to interpret this as a wheel and to find any defects it may have.) However, an exception to this general strategy is the relational descriptor analysis of scenes that appears toward the end of Chapter 15: this approach may be regarded as constituting the foothills of “high-level” vision.