Page 313 - Fundamentals of Light Microscopy and Electronic Imaging

296 DIGITAL IMAGE PROCESSING

called histogram equalization or histogram leveling, yields a histogram that contains no
peaks and has a flat, horizontal profile. In leveling, pixel values are reassigned so that
each gray level is given the same number of pixels, while the rank order of the pixel val-
ues in the original picture is preserved as much as possible. This operation is used to
enhance contrast in very-low-contrast (flat) images where most of the pixels have close
to the same value, and where conventional methods of histogram stretching are ineffec-
tive. Equalization is an extreme measure to rescue images with low-amplitude gradients,
but works well as a way to examine bias, dark, and flat-field frames, which can look
nearly featureless. The effect of histogram equalization is often dramatic.

Unsharp Masking

This image-processing procedure does an excellent job of enhancing fine details in an
image. Unsharp masking is well known to photographers and astronomers who used the
method as a darkroom technique to enhance faint details in photographic prints. As
designed by photographers, a blurred, reverse-contrast negative (or unsharp mask) is
made of the original negative. The two negatives are sandwiched together in perfect reg-
istration in the enlarger and a print is made. To perform this operation on a computer, an
unsharp mask is produced by blurring and reducing the amplitude of the original image;
the unsharp mask is then subtracted from the original to produce a sharpened image.
Extended, uniform regions are rendered a medium gray, whereas regions with sharp gra-
dients in amplitude appear as brighter or darker intensities. The principle of unsharp
masking is shown in Figure 15-9, and an example of a microscope image processed by
this method is shown in Figure 15-10. If the program you are using does not perform
unsharp masking, you can perform this operation manually using the following steps:

• Prepare a copy of the original and blur it with a conservative blurring filter. Images
with fine details (high spatial frequencies) require more conservative blurring than
images containing big blocky objects (low spatial frequencies).
• Subtract 50–95% of the amplitude of the blurred image from 100% of the original
using an image math function in the program. The higher the percentage that is sub-
tracted, the greater the sharpening effect.
• Using histogram stretching, adjust the brightness and contrast in the difference
image.

Fast Fourier Transform (FFT)

This filtering operation selectively diminishes or enhances low or high spatial frequen-
cies (extended vs. fine detailed structures) in the object image. This is a valuable opera-
tion to consider, because it reinforces concepts given in previous chapters on the
location of high- and low-spatial-frequency information in the diffraction plane, which
is located in the objective back aperture in the microscope. The effect is similar to that
of the blurring and sharpening filters already described, but can be made to be much
more specific due to an operation called spatial frequency filtering (Fig. 15-11). When
an image is transformed into the so-called frequency domain through an FFT command,
the information is represented in two plots (images): one containing a distribution of