Page 244 - Digital Analysis of Remotely Sensed Imagery

208 Cha pte r S i x

In the absence of these critical points, DN and DN may be defined
min max
statistically, such as 5 percent of the total number of pixels in either
side of the tail, or the mean DN ± one standard deviation. Defined in
such a way, DN is larger than the actual minimum DN, while DN
min max
is smaller than the actual maximum DN. Both are less subject to ran-
dom noise than the actual DN and DN . In truncated stretching, all
min max
the input pixel values smaller than this statistical DN are assigned
min
the same minimum value, whereas all input pixel values above this
statistical DN receive the same maximum value in the output image
max
(Fig. 6.3c), resulting in a higher stretching ratio than full stretching. All
the pixel values lying between these two extremes are stretched lin-
early just as in nontruncated stretching. Inevitably, truncated linear
stretching involves loss of information at both tails of the distribution.
Compared with the raw image (Fig. 6.4a), any subtle radiometric varia-
tions outside the DN − DN range are generalized in the output,
min max
which exhibits more detail thanks to its enhanced quality (Fig. 6.4b).

6.1.3 Piecewise Linear Enhancement
The contrast of the same input image may be linearly stretched differ-
ently for different pixel values in a piecemeal manner (Fig. 6.5).
Instead of a single stretching function f for all DNs, a few linear func-
tions are used for the stretching. Each function segment has its own
slope and is applicable to a specific range of digital numbers. This is
known as piecemeal linear enhancement. With the use of multiple
enhancement functions, it is possible to stretch the contrast of an
image at different pixel values. For instance, the contrast within a
certain range of DNs is artificially enlarged just as in ordinary linear
stretching, but the contrast over another DN range is suppressed.
Suppression of contrast over a DN range that falls outside the scope
of interest leaves more room to stretch the contrast over a wider range
of DNs for features of interest (e.g., water turbidity). Through sacri-
ficing the information of uninteresting features, features of interest
are rendered more prominently in a piecewise linearly stretched
image than a single linear stretching.

6.1.4 Look-Up Table
Look-up table is a method of adjusting the value of pixels in an input
image based on a purposely defined scheme. Contained in this
scheme is a series of arbitrarily but deliberately designed values cor-
responding to every potential value in the input band. A look-up
table is an effective way of visualizing an image. If it is black and
white, it can be easily rendered as a gray image, using only one series
of numbers. However, three series of numbers are needed for its color
rendition. In each series of numbers, there is a unique correspondence
between an input value and the designated output value. Look-up