Image Analysis, Classification and Change Detection in Remote Sensing, 3rd Edition

High-pass spatial filtering and feature extraction 169

FIGURE 5.6

Corners and edges detected from a small spatial subset of the ASTER image

of Figure 1.1. Center: a subse t of the first sp ectral band. Left: the minimum

eigenvalue intensities in a linear 2% histogram stretch. Right: the Canny

edges.

5.2.3.2 Canny edge detector

The Ca nny edge detecto r (Canny, 1986),

∗

which is similar to the corner de-

tection algorithm, is based upon a gradient filter such as the Sobel filter.

First a Gaus sian smoothing filter is a pplied to suppress noise , followed by the

gradient filter, whereby both the magnitude of the gradient

k∇g(i, j))k =

p

∇

1

(x)

2

+ ∇

2

(x)

2

as well as