Image segmentation is a process of separating objects of interest from an image background and is of
a crucial preprocessing step for most object recognition systems. In general, the accuracy of classifica-
tion/recognition system depends heavily on the accuracy of object features used in a training process.
More precise segmentation result contributes to more accurate object feature computation.
The main difficulties of segmenting algae from an image background are noise and a blurred contour
and texture as discussed earlier. Most microscopic images of algae are usually corrupted by noise. Noise
in an image can be extraneous materials (or unwanted objects) and illumination artefacts. These noise
disrupt a segmentation process and it is not trivial to remove them without a loss of object information.
Moreover, it is often to occur that noise have similar characteristics to objects of interest. Thus, it is
quite problematic to a computer to automatically distinguish them by considering their features.
One of the most powerful tools for noise suppression is image smoothing (also known as lowpass
filtering). Image smoothing suppresses the noise by attenuating its signal which makes its intensity
roughly consistent with those of its nearest neighbors. Unfortunately, in many cases, i) polluted objects
are much clearer and sharper than spines (in Scenedesmus, Xanthidium, and Staurastrum genera) and
flagellums (in Euglena and Phacus genera) of algae; and ii) a thick gelatinous coat of algae is sharper
than a true algae boundary. If we perform a high degree of noise suppression in order to remove polluted
objects and a gelatinous coat, this usually removes spines, flagellums, and internode contours of these
algae. On the other hand, if we perform a low degree of noise suppression, the detected boundary of
algae body often distorts and lies further away from the true boundary of the algae (due to touching
polluted objects and a thick gelatinous coat of algae) .
This situation causes a serious problem to classifying algae in Anabaena, Oscillatoria and Pleuro-
taenium genera. The algae in these genera have similar rod shape. The main difference between their
shapes is that algae in Oscillatoria and Pleurotaenium genera have smooth boundary, while algae in
Anabaena genus have ripple along its boundary. If we perform insufficient image smoothing, the ripple
along the boundary of algae in Anabaena genus disappears (due to gelatinous coat), and the smooth
boundary of the algae in Oscillatoria and Pleurotaenium genera becomes ripple (due to small touching
polluted objects).
Image segmentation is a process of separating objects of interest from an image background and is of
a crucial preprocessing step for most object recognition systems. In general, the accuracy of classifica-
tion/recognition system depends heavily on the accuracy of object features used in a training process.
More precise segmentation result contributes to more accurate object feature computation.
The main difficulties of segmenting algae from an image background are noise and a blurred contour
and texture as discussed earlier. Most microscopic images of algae are usually corrupted by noise. Noise
in an image can be extraneous materials (or unwanted objects) and illumination artefacts. These noise
disrupt a segmentation process and it is not trivial to remove them without a loss of object information.
Moreover, it is often to occur that noise have similar characteristics to objects of interest. Thus, it is
quite problematic to a computer to automatically distinguish them by considering their features.
One of the most powerful tools for noise suppression is image smoothing (also known as lowpass
filtering). Image smoothing suppresses the noise by attenuating its signal which makes its intensity
roughly consistent with those of its nearest neighbors. Unfortunately, in many cases, i) polluted objects
are much clearer and sharper than spines (in Scenedesmus, Xanthidium, and Staurastrum genera) and
flagellums (in Euglena and Phacus genera) of algae; and ii) a thick gelatinous coat of algae is sharper
than a true algae boundary. If we perform a high degree of noise suppression in order to remove polluted
objects and a gelatinous coat, this usually removes spines, flagellums, and internode contours of these
algae. On the other hand, if we perform a low degree of noise suppression, the detected boundary of
algae body often distorts and lies further away from the true boundary of the algae (due to touching
polluted objects and a thick gelatinous coat of algae) .
This situation causes a serious problem to classifying algae in Anabaena, Oscillatoria and Pleuro-
taenium genera. The algae in these genera have similar rod shape. The main difference between their
shapes is that algae in Oscillatoria and Pleurotaenium genera have smooth boundary, while algae in
Anabaena genus have ripple along its boundary. If we perform insufficient image smoothing, the ripple
along the boundary of algae in Anabaena genus disappears (due to gelatinous coat), and the smooth
boundary of the algae in Oscillatoria and Pleurotaenium genera becomes ripple (due to small touching
polluted objects).
การแปล กรุณารอสักครู่..
![](//thimg.ilovetranslation.com/pic/loading_3.gif?v=b9814dd30c1d7c59_8619)