4 - Low-level feature extraction (including edge detection)

Abstract

We shall define low-level features to be those basic features that can be extracted automatically from an image without any shape information (information about spatial relationships). As such, thresholding is actually a form of low-level feature extraction performed as a point operation. Naturally, all of these approaches can be used in high-level feature extraction, where we find shapes in images. There are very basic techniques and more advanced ones and we shall look at some of the most popular approaches. The first order detectors are equivalent to first order differentiation and, naturally, the second order edge detection operators are equivalent to a one-higher level of differentiation. An alternative form of edge detection is called phase congruency and we shall again see the frequency domain used to aid analysis, this time for low-level feature extraction. We shall also consider corner detection which can be thought of as detecting those points where lines bend very sharply with high curvature, and saliency which are important points. These are another low-level features that again can be extracted automatically from the image. Finally, we shall investigate techniques that describes motion, called optical flow.