An Introduction to Remotely Sensed Data Analysis - Sampling Spatial Units for Agricultural Surveys - page 76

changing the numerical values of the original image according to some specific
function. The new histogram will have the same number of bars, but they will be
located on different values. Define the distribution of brightness values associated
with contrast stretching as

y ¼ fx ðÞ; ð 4 : 11 Þ

where x is the old brightness value of a particular bar, and y is the corresponding
new brightness value. Our aim is to define a functional form for Eq. ( 4.11 ). The
most common choice for f ( x ) is a simple linear model. Then, Eq. ( 4.11 ) can be
expressed as

y ¼ a þ bx : ð 4 : 12 Þ

The parameters a and b are usually chosen in a way that the minimum and
maximum brightness values in the original image are expanded to the lowest and
highest brightness levels supported by the display device. Other definitions for f ( x )
are possible. Logarithmic functions are very useful for enhancing dark features, and
exponential functions similarly enhance light features.
Histogram equalization is another method that is widely used in image enhance-
ment. This technique redistributes pixel intensity values to spread (evenly distrib-
ute) the image histogram, increasing the dynamic range by increasing the image
contrast. One possible transformation function for this is a cumulative distribution
function (cdf). The method is useful for images with backgrounds and foregrounds
that are both bright or both dark. It tends to reveal details that would otherwise be
hidden. It often produces unrealistic effects in photographs, but is very useful in
scientific images such as X-ray, satellite, or thermal images. Histogram equaliza-
tion differs from contrast stretching in that it uses non-linear transformation func-
tions to map the input pixel intensity values to the output images.
Spatial filtering is a pixel-by-pixel transformation of the image, and is designed
to improve an image

s readability. The principle behind the various filters is very
simple, it consists of modifying the numerical value of each pixel using a function
of the neighboring pixel values. Here, the aim is to locally expand the contrast in the
spatial domain.
Often, the pixel brightness can contain a random noise generated from the
acquisition process of the sensors. In this case, the noise can be removed using
low pass filtering. For example, if the value of each pixel is replaced by the average
of its eight neighbors the image is smoothed, the finer details disappear, and the
image looks fuzzier. It is obvious that high frequency information such as edges
will also be averaged and consequently lost. This last problem can be overcome if a
threshold procedure is applied (Richards and Jia 2006 ).
Edge detection techniques can be used to define the boundaries in an image
(Nadernejad et al. 2008 ). Edge detection is a very important area in the field of
computer vision. An edge is defined as significant local changes of intensity in an
image. Edge enhancement is performed by first detecting edges, and then adding

'