Geography Reference
In-Depth Information
or stocks, to characterise life-history traits, to describe
behaviours, to detect diseases, to make chronic stress
diagnostics or to relate morphology and ecology.
This chapter presents selected examples of how imagery
techniques are currently (or could be) used in freshwater
studies at the level of the organism.
17.2 Morphological and anatomical
description
17.2.1 Identification
Automated (or semi-automated) systems of shape iden-
tification with IA almost always follow the same sequence
of procedures: (1) background elimination, (2) image
segmentation (i.e. location of objects of interest and often
shift to a binary image, Figure 17.2), (3) focus check,
(4) object feature extraction, (5) feature selection and
measurement, (6) feature analysis, (7) classification and
(8) estimation of system performance. Such protocols
were widely used to recognise and classify various taxa,
but other techniques (e.g. recognition of colouration
patterns) can also be very effective.
Figure 17.1
Morphometric variation of fishes, expressed as
regular deformations of rectangular coordinate systems.
Reproduced from Thompson, D.W. (1917)
On Growth and
Form
, with permission from Cambridge University Press.
17.2.1.1 Species (or taxa) recognition
Although automated species recognition has not replaced
experts (see why in Gaston and O'Neill, 2004), numerous
studies developed routine IA identification, for various
purposes.
Bacterial morphological diversity could be an indica-
tor of dynamic ecological succession following a nutrient
perturbation in bacterial communities (Liu et al., 2001).
In microbial ecology, a major challenge is to develop reli-
able methods of computer assisted microscopy that can
analyse digital images of complex microbial communities
at the resolution of the single cell, and to compute useful
quantitative characteristics of their organisation with-
out the need for cultivation. Liu et al. (2001) described
a computer-aided system (Figure 17.3), which extracted
size and shape measurements of segmented, digital images
of microorganisms and classified them into one of 11
predominant bacterial morphotypes (e.g. cocci, spirals,
curved rods, ellipsoids). This shape classifier had an accu-
racy of 97% on a test set of 4,270 cells representing
all these bacterial morphotypes, indicating that accurate
classification of rich morphological diversity in microbial
communities was possible using IA.
To complement shape recognition, bacterial cells can
be classified in a given taxa (or group) after specific
can be visualised through Thin Plate Spline (TPS) analysis
(see Zelditch et al., 2004 and Figure 17.1).
Shape characterisation based on landmarks has pro-
duced valuable results; however, the main difficulty of
this approach is that landmark points cannot be located
very accurately all of the time. In this case, other meth-
ods, such as outline analyses, can be used. Most of these
methods consist in expressing outlines in periodic sig-
nals. Using Fourier transform, such signals are fitted by a
sum of trigonometric functions (or harmonics) that have
different amplitudes and phases (see a review of shape
descriptors in Zhang and Lu, 2004).
Imagery methods are currently applied, for various
purposes, on almost all biological model organisms (from
viruses to vertebrates). Observations can be automated
and are sometimes made
in situ
. At the species (or multi-
species) level, imagery was used to detect the presence of
specimens and/or estimate their body lengths, abundance
and biomass; methods were also developed for recognis-
ing and separating various species from each other. At
the infra-species (i.e. within species or sub-species) level,
IA has helped to differentiate sexes, ontogenetic stages
Search WWH ::
Custom Search