
obtained from a blank image field. This image, called
the shading pattern, is used to correct subsequent
images.
0025 Image processing can never increase the informa-
tion content of an image. Any enhancement of
one image attribute is obtained at the expense of
another. For example, edge enhancement algorithms
tend to increase the noise content of an image. Many
algorithms have been developed over the years to
enhance images with the minimum of unwelcome
side-effects. There are, for example, methods of
reducing noise without smoothing out the edges of
objects. However, heavy processing can still intro-
duce unwanted artifacts in the image. For this reason
sample preparation and lighting conditions are best
chosen to reduce the need for image enhancement to a
minimum.
Segmentation
0026 This is the process by which the digitized, gray,
or color image is converted into a binary image
representing the objects to be measured. This is the
most important, and often the most difficult, step in
image analysis. The computer needs a binary image
to make measurements. There are several ways of
defining the regions to be segmented. The simplest
form is by setting a gray-level value below which
(or above, for light objects) a pixel will be counted
as being part of the object. All other pixels will
be counted as belonging to the background. This
form of segmentation is also referred to as threshold-
ing or detection. Another method allows the operator
to draw round or point to a typical region; the com-
puter will sample the intensities of the pixels in the
region and generate a binary image of all regions
containing pixels with matching intensities. There
are also methods that specifically detect the edges of
objects.
0027 Color discrimination provides an alternative
method of segmentation for true color images. There
are two main color models used in image analysis–the
red, green, blue (RGB) model and the hue, luminance,
saturation (HLS) model. Color images are separated
into their three components (RGB or HLS) according
to which color model is to be used for segmentation.
Each component is held in memory as a separate
image. These component images can be viewed separ-
ately as monochrome images but are normally dis-
played combined as a color image (Figure 2).
Segmentation methods, similar to those used on
monochrome images, are applied to each of these
components and the resultant binary images com-
bined. The resultant binary image will represent
those regions in the original image that have a similar
color composition.
Binary Image Processing
0028Often the segmentation process does not produce a
perfect binary image of the objects of interest and
many binary operations have been devised to amend
the defects in the image. For example, there are oper-
ations for filling in holes and others that dilate or
erode the binary image by adding or subtracting
pixels from the perimeter of objects. Erosion and
dilation can often be used to good effect by combin-
ing them to form new operations known as closing
and opening. Closing is the process of dilation
followed by an equal amount of erosion and is useful
for joining up fragmented regions of an object.
Opening is the converse operation and is often used
to clean up ragged edges and remove small or thin
artifacts from the image. Opening and closing oper-
ations leave the overall size of the objects relatively
unchanged compared with erosion or dilation used on
their own.
0029There are other more complex binary operations
which can be used, for example, to separate touching
objects, reduce objects to a skeletal representation,
and erode objects to the point just before they
would otherwise disappear. These tools allow the
image analyst to automate much of the editing of
binary images that would otherwise have to be done
by time-consuming manual methods. They form an
important part of image analysis and can make the
difference between being able or otherwise to per-
form an analysis in a reasonable time. The theory
behind these operations is known as mathematical
morphology.
Editing
0030The binary image can be edited manually by using a
hand-held device, such as a mouse or a light pen,
which is interactive with the computer. By this
means sections or lines can be cut out or drawn in at
will. Although it is often necessary to do a little
manual editing, this should always be regarded as a
method of last resort since it is a slow operation and
introduces a subjective element into the analysis.
Measurement
0031There are two distinct modes of measurement that can
be made in image analysis: field-specific and object-
specific. Before discussing these it is important to dis-
tinguish between the image and measurement frames.
The total area of the specimen scanned by the TV
camera and displayed on the monitor forms the image
frame. The measurement frame is a subframe within
the image frame over which the measurements are
made. The area within the image frame that lies outside
the measurement frame forms the guard region.
MICROSCOPY/Image Analysis 3937