Geoscience Reference
In-Depth Information
I5
an intensity image with
n
discrete levels. A roughly equal number of
pixels is ascribed to each of the
n
levels in
I5
, so that the histogram of
I5
is approximately l at. Histogram equalization using
histeq
has to carried
out separately for each color, since
histeq
only works for two-dimensional
images. We use
n=50
in our exercise, which is slightly below the default value
of
n=64
.
I5(:,:,1) = histeq(I1(:,:,1),50);
I5(:,:,2) = histeq(I1(:,:,2),50);
I5(:,:,3) = histeq(I1(:,:,3),50);
subplot(2,2,1), imshow(I1), title('Original Image')
subplot(2,2,3), imhist(I1(:,:,1)), title('Original Image')
subplot(2,2,2), imshow(I5), title('Enhanced Image')
subplot(2,2,4), imhist(I5(:,:,1)), title('Enhanced Image')
h e resulting image looks quite disappointing and we therefore use the
improved function
adapthisteq
instead of
histeq
. h is function uses the
contrast-limited adaptive histogram equalization (CLAHE) by Zuiderveld
(1994). Unlike
histeq
and
imadjust
, the algorithm works on small regions (or
tiles) of the image, rather than on the entire image. h e neighboring tiles are
then combined using bilinear interpolation to eliminate edge ef ects.
I6(:,:,1) = adapthisteq(I1(:,:,1));
I6(:,:,2) = adapthisteq(I1(:,:,2));
I6(:,:,3) = adapthisteq(I1(:,:,3));
subplot(2,2,1), imshow(I1), title('Original Image')
subplot(2,2,3), imhist(I1(:,:,1)), title('Original Image')
subplot(2,2,2), imshow(I6), title('Enhanced Image')
subplot(2,2,4), imhist(I6(:,:,1)), title('Enhanced Image')
h e result looks slightly better than that obtained using
histeq
. However, all
three functions for image enhancement,
imadjust
,
histeq
and
adapthisteq
,
provide numerous ways to manipulate the i nal outcome. h e
Image
Processing Toolbox - User
'
s Guide
(MathWorks 2014) and the excellent topic
by Gonzalez and others (2009) provide more detailed introductions to the
use of the various parameters available and the corresponding values of the
image enhancement functions.
h e Image Processing Toolbox also includes numerous functions for 2D
i ltering of images. Many of the methods we have looked at in Chapter 6
for one-dimensional data also work with two-dimensional data, as we have
already seen in Chapter 7 when i ltering digital terrain models. h e most
popular 2D i lters for images are Gaussian i lters and median i lters, as well
as i lters for image sharpening. Both Gaussian and median i lters are used
to smooth an image, mostly with the aim of reducing the amount of noise.