Page 38 - Digital Analysis of Remotely Sensed Imagery

Overview 11

Since mixed pixels do not have a singular identity, it is impossible to
correctly classify them into any one component cover at the pixel level.
Their precise labeling has to take place at the subpixel level with a
probability attached to each component feature. No matter whether a
pixel is pure or mixed, it always has two crucial properties, its value or
digital number (DN), and its location in a two-dimensional space.

1.4.2 Digital Number (DN)
The DN of a pixel in a spectral band represents the amount of radiation
received at the sensor, which is determined primarily by the capability
of the ground object in reflecting and emitting energy. The amount of
energy reaching the sensor is a function of the wavelength of the
radiation. Thus, pixel value varies from band to band. The actual DN
value of a pixel in an image is affected by many other external factors,
such as atmospheric radiation, the sensor’s sensitivity, and more
importantly, the ground sampling interval of the sensing system. In
spite of these external interferences, theoretically, the same target
should have the same or similar DN value in the same band; and
different targets should have dissimilar DN values in the same band.
However, this relationship is not always maintained because of the
similar appearance of some ground objects.
No matter how many bands the received energy is split into
spectrally, it is always recorded as positive integers (Fig. 1.3). The
theoretical range of pixel values in an image is determined by the
number of bites used to record the energy, or the quantization level.
A commonly adopted quantization level is 8 bits. So the number of
8
potential DN values amounts to 2 or 256, ranging from 0 to 255. A
DN value of 0 implies that no radiative energy is received from the
target on the ground. A value of 255 indicates a huge amount of
radiation has reached the sensor in space. Because of the atmospheric
impact or limitations in the sensing system, not all of the potential
levels of DN are fully taken advantage of during data recording, a
situation that can be remedied through image enhancement. Recent
advances in sensing technology have made it possible to reach a
quantization level as high as 11 bits. As illustrated in Fig. 1.3, at a
9
quantization level of 9, pixel values vary from 0 to 2 − 1 (511). In the
binary system of encoding the amount of received energy, pixel
values are not allowed to have any decimal points. They are recorded
as 8-bit, unsigned integers. Floating point pixel values are not
commonly associated with raw satellite data. With the use of more
bits in a computer, it is possible to have floating point data for some
processed results (e.g., ratioed band).
1.4.3 Image Reference System
There are many coordinate systems in use, such as the latitude-
longitude system and the cartesian coordinate system. The latter is
a plane system suitable for representing two-dimensional digital