Page 240 - Digital Analysis of Remotely Sensed Imagery

204 Cha pte r S i x

transformation in detail, its main use, and interpretation of the trans-
formed results. Finally, this chapter ends with a section on image
filtering in the frequency domain. In all discussions, mathematical
equations and calculations are provided for those readers with the
necessary background. Those readers who are not interested in the
mathematical underpinning of image enhancement may choose to
focus on the interpretation of the transformed results.

6.1 Contrast Stretching
Contrast stretching is a process of modifying or enlarging the range
of pixel values in an input image in an attempt to improve its visual
effectiveness or quality. In this process the digital number (DN) value
of every pixel in the image is modified according to a pre-determined
function. It includes density slicing and contrast enhancement, both
carried out for single bands. In this histogram-based operation, a
pixel’s DN is modified regardless of its neighboring pixels’ values.
Mathematically, contrast stretching is expressed as

DN = f(DN ) (6.1)
out in
where DN = output DN in the contrast-stretched image
out
DN = DN of the same pixel in the raw image
in
f = transformation function through which contrast is
manipulated; it can be either linear or nonlinear

6.1.1 Density Slicing
Also known as pixel-value thresholding, density slicing is virtually a
process of discretizing the continuously varying pixel values in the
input band. Pixel values within a certain gray level range are amalgam-
ated into a single value in the output image. The range of entire pixel
values in the input image is reduced to a few categories of values, each
corresponding to a unique range of pixel values in the input image.
Thus, the potential number of pixel values is considerably reduced in
the sliced image. A unique color may then be assigned to each newly
created pixel value, converting a gray level image into a pseudocolor
one. Since the naked human eyes are much more sensitive to variation
in color than gray tone, the subtle spatial pattern contained in the input
image is much more easily perceived visually in the density-sliced out-
put. In order to produce a meaningful pattern for the phenomenon
under study (e.g., concentration levels of silt in nearshore water), the
thresholds for each discrete category must be carefully selected (Fig. 6.1).
Frequently, the histogram of a spectral band is relied to derive the criti-
cal slicing thresholds that should not be overlapping across categories.
The more appropriately selected these thresholds are, the more authen-
tic the resultant spatial pattern is.