1 October 2003
(5) Histogram Equalization. Low contrast can also occur when values are spread
across the entire range. The low contrast is a result of tight clustering of pixels in one
area (Figure 5-10a). Because some pixel values span the intensity range it is not possible
to apply the contrast linear stretch. In Figure 5-10a, the high peak on the low intensity
end of the histogram indicates that a narrow range of DNs is used by a large number of
pixels. This explains why the image appears dark despite the span of values across the
full 0255 range.
(a) Histogram equalization evenly distributes the pixel values over the entire
intensity range (see steps below). The pixels in a scene are numerically arranged ac-
cording to their DN values and divided into 255 equal-sized groups. The lowest level is
assigned a gray level of zero, the next group is assigned DN 1, ..., the highest group is
assigned gray level 255. If a single DN value has more pixels than a group, gray levels
will be skipped. This produces gaps in the histogram distribution. The resultant shape of
the graph will depend on the frequency of the scene.
(b) This method generally reduces the peaks in the histogram, resulting in a
flatter or lower curve (Figure 5-10b). The histogram equalization method tends to en-
hance distinctions within the darkest and brightest pixels, sacrificing distinctions in mid-
dle-gray. This process will result in an overall increase in image contrast (Figure 5-10b).
(6) Logarithmic Enhancement. Another type of enhancement stretch uses a loga-
rithmic algorithm. This type of enhancement distinguishes lower DN values. The high
intensity values are grouped together, which sacrifices the distinction of pixels with
(7) Manual Enhancement. Some software packages will allow users to define an
arbitrary enhancement. This can be done graphically or numerically. Manually adjusting
the enhancement allows the user to reduce the signal noise in addition to reducing the
contrast in unimportant pixels. Note: The processes described above do not alter the
spectral radiance of the pixel raw data. Instead, the output display of the radiance is
modified by a computed algorithm to improve image quality.
b. Image Enhancement #2: Band Arithmetic
(1) Band Arithmetic. Spectral band data values can be combined using arithmetic
to create a new "band." The digital number values can be summed, subtracted, multi-
plied, and divided (see equations 5-1 and 5-2). Image software easily performs these op-
erations. This section will review only those arithmetic processes that involve the divi-
sion or ratio of digital band data.
(2) Band Ratio. Band ratio is a commonly used band arithmetic method in which
one spectral band is proportioned with another spectral band. This simple method re-
duces the effects of shadowing caused by topography, highlights particular image ele-
ments, and accentuates temporal differences (Figure 5-11).