Low-level feature extraction (including edge detection) - Feature Extraction and Image Processing - page 99

4

Low-level feature
extraction (including
edge detection)

4.1 Overview

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
shown in Table 4.1. 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. It is well known that we can recognise
people from caricaturists' portraits. That is the first low-level feature we shall encounter. It
is called edge detection and it aims to produce a line drawing , like one of a face in Figures
4.1 (a) and (d), something akin to a caricaturist's sketch though without the exaggeration a
caricaturist would imbue. 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.
We shall also consider corner detection which can be thought of as detecting those
points where lines bend very sharply with high curvature , as for the aeroplane in Figures
4.1 (b) and (e). These are another low-level feature that again can be extracted automatically
from the image. Finally, we shall investigate a technique that describes motion , called
optical flow. This is illustrated in Figures 4.1 (c) and (f) with the optical flow from images
of a walking man: the bits that are moving fastest are the brightest points, like the hands
and the feet. All of these can provide a set of points, albeit points with different properties,
but all are suitable for grouping for shape extraction. Consider a square box moving though
a sequence of images. The edges are the perimeter of the box; the corners are the apices;
the flow is how the box moves. All these can be collected together to find the moving box.
We shall start with the edge detection techniques, with the first-order operators which
accord with the chronology of development. The first-order techniques date back more
than 30 years.

4.2 First-order edge detection operators

4.2.1 Basic operators
Many approaches to image interpretation are based on edges, since analysis based on edge