Geography Reference
In-Depth Information
14000
12000
10000
8000
6000
4000
2000
0
0
50
100
Bin (DN)
150
200
250
(a)
(b)
(c)
2.5 × 10 4
2
1.5
1
0.5
0
0
50
100
Bin (DN)
150
200
250
20 cm
(f)
(d)
(e)
Figure 9.2 Surficial sand identification based on normal image brightness values (a, b, c) and textural entropy values (d, e, f). In
a and d, pixel brightness is respectively proportional to reflected radiation and neighbourhood image texture. b and e are the
histograms of images a and d respectively. c and f show the results of the binary segmentation process obtained respectively from the
radiation (b) and texture (e) histograms. Reproduced from Carbonneau et al. (2005b), with permission fromWiley-Blackwell.
this colour within a local spatial neighbourhood. While
a simple metric such as local variance could be used to
quantify local colour variability, the concept of image
texture is most often employed. Image texture is defined
as: 'an attribute representing the spatial arrangement of
the grey levels of the pixels in a region' (Haralick and
Shapiro, 1985). This texture property therefore allows
the production of a new image where the brightness of
each pixel is in fact proportional to the texture of a
given neighbourhood instead of being proportional to
the reflected radiation. The result of this process is shown
in Figure 9.2d. Here it can be noticed that the perceived
brightness difference between the sand patch and the clast
has been greatly amplified. As a result, the histogram in
Figure 9.2e shows a much greater separation between
the two modes. Consequently, the resulting segmentation
shown in Figure 9.2f is now much less speckled and is
a more accurate reflection of reality. After testing on a
range of 20 images, Carbonneau et al. (2005b) found that
the texture based classification predicted values of sand
coverage in terms of a surface % of the whole image,
with an R 2 of 93% versus only 56% for traditional grey
level brightness segmentation. Therefore, this approach
can be effectively used during walkover surveys. All that is
needed is an image with a known scale, and sand coverage
can be sampled. Given that a digital picture takes seconds
to acquire, this approach is very cost effective in the field.
Furthermore, the approach can be combined to dubbed
photosieving methods that use terrestrial images in order
to measure particle sizes (Ibbeken and Schleyer, 1986;
Dugdale et al., 2010). However it should be noted that
this method is best used at periods of very low flow when
significant areas of river bed are dry and exposed. In the
case of submerged bed material, the image histogram is
compressed in a manner which is similar to the shading
effect discussed in Chapter 8 (see Figure 9.12 of that
Search WWH ::




Custom Search