Page 126 - Digital Analysis of Remotely Sensed Imagery

Storage of Remotely Sensed Data 97

• Second, each reduced value is recoded, starting from the
smallest to the original one, in a backward sequence. The two
remaining probabilities are assigned the simplest codes of 0
and 1 arbitrarily to distinguish them from each other. If 0 is
assigned to the value of a lower probability, it has to remain
unchanged in subsequent operations in which more codes
are added to the front of the existing code(s).
• Third, repeat the process until the original source is reached.
All codes for the source values end up having a varying
length. The original pixel values are decoded by examining
the string of codes from left to right in a lookup table manner.
This string of codes is uniquely decodable.

For 8-bit remote sensing imagery, there are 256 possible pixel
values. The construction of the optimal binary Huffman code (i.e.,
254 source reductions and 254 code assignments) is a daunting task.
It is hence necessary to simplify code construction at the expense of
coding efficiency.
In lossless predictive coding, interpixel redundancies of closely
spaced pixels are eliminated through extraction and coding of only
the difference between the actual and predicted value in each pixel.
The degree of data compression achievable with this method is related
directly to the entropy reduction.

3.4.2 Run-Length Coding
It is quite common to encounter a long string of the same pixel value
in a row, especially if the image is binary. Binary images can be created
through a process called bit-plane coding. An original image of either
grayscale or color is decomposed into binary ones in the form of the
base 2 polynomial. For instance, the tone of an image changes little
across the same hayfield. In this case, the run-length coding method
can be adopted to reduce image data. In this coding, two values are
used to code a particular value, say zero. Whenever this value is
encountered in a row from left to right, it is coded by two values in
the output file. The first is the value of the first run, and the second
describes the number of times it is encountered or successive runs
(Fig. 3.7). Thus, a high efficiency of data compression is expected if a
long contiguous string of pixels have the same value. The input data
can be treated as individual bytes, or groups of bytes that represent
something more elaborate, such as floating point numbers. Although
this method is an effective means of data compression, run-length

Original data stream: 25 34 4 4 4 32 28 34 4

Run-length codes: 25 1 34 1 4 3 32 1 28 1 34 1 4 1
FIGURE 3.7 An example of run-length coding.