Image Analysis: Basic Procedures for Description of Plant Structures - Plant Cell Morphogenesis: Methods and Protocols - page 67

Objects in the plane perpendicular to the objective axis could
be acquired without distortion ( see Note 1 ) and their physical
properties might be analyzed, if the size of the image pixels is
known. Dimensions of pixels are determined during calibration
( see Note 2 ).
Selection of proper image format affects quality of the image
data. Some formats (e.g., JPG) use lossy compression and might
cause some data degradation. That is why lossless formats (e.g.,
TIFF) are far more suitable for primary data storage. The images
intended for image analysis should have suffi cient both image reso-
lution and bit depth. The size of the pixel shall be at least one half
of the least requested detail in the image as defi ned by the Nyquist
sampling theorem. Higher bit depth (12 bits and more) enables
more accurate image processing and segmentation.
Image preprocessing can be used to correct some defi ciencies
(background gradients, dust, etc.) and to improve visual image qual-
ity. Unlike human the computer does not identify the object of inter-
est, so image segmentation should be used to separate those from
background. The most common and simplest method of segmenta-
tion is thresholding, which results in binary image (binary values per
pixel; 0/1; black/white) defi nitions of the objects. Subsequent
adjustments of binary image based on mathematic morphology—
operations of dilation, erosion, closing, and opening—are used to
improve object representation. Objects, once defi ned, can be mea-
sured and classifi ed using various geometrical parameters ( see Note
3 ). Subset of objects might be selected in specifi ed region of interest
(ROI), with superimposed mask or a counting frame [ 1 - 3 ].
The majority of plant structures exhibits gradients [ 4 ] in quan-
titative anatomical parameters, refl ecting polarity of plant structures
established during early plant ontogenetic development and organ-
ogenesis. A known example is the dependence of some anatomical
parameters of leaf on its insertion—i.e., distance from the root sys-
tem. Because of the spatial heterogeneity of values of anatomical
parameters in biological structures, it might be diffi cult to specify
average values of a particular anatomical parameters within an organ
or tissue. Proper sampling designs based on a systematic uniform
random (SUR) sampling (e.g., ref. [ 2 ]) are necessary to gain repro-
ducible values. Unbiased estimation of a measured parameter can
be achieved by SUR sampling, which ensures the prerequisite of
unbiased estimation, i.e., that each particle or object has the same
probability to be selected by the sampling system or rule. This is a
very important issue, often neglected in formation of sampling
design of biological studies. Quite often the sampling design and
the method of quantifi cation are not fully described, and thus the
results can be diffi cult to compare and discuss with other results as
a consequence of biological treatment or processes.
Stereological methods might be also employed to analyze an
image. Stereological methods provide 3D characteristics of objects
based on measurement of parameters in 2D or 3D image, which is