Environmental Engineering Reference
In-Depth Information
Original image
Score density histogram
t
2
PCA
t
1
Masking
Extracted features
t
2
Pixel
identification
t
1
Figure 3.7
Example of MIA applied to a froth flotation image. The blue and green masks identify
pixels belonging to brown areas (
i.e.
, sphalerite) and clear windows (appear as black), respectively
the minimum and maximum values of all
t
1
and
t
2
score vectors of the set of images
(
i.e.
, training set). Such a scaling range may be exceeded when the model is used
on-line and new operating conditions, falling outside the operating range included
in the training set, are experienced. This situation can, however, easily be monitored
and, when detected, new conditions may be incorporated in the model through some
updating strategy.
The spectral features extracted from the set of images using MIA can be for-
mulated in various ways depending on the degree of complexity of the features of
interest. This was investigated by Yu and MacGregor [32] who classified the fea-
tures in two categories: overall and distribution features. If one is only interested in
tracking overall color variations from one image to the other, then the weights of the
loading vectors
p
a
obtained from each individual image
X
j
(not from
Z
) can be used
as representative color features of the
j
th image for further analysis. If, on the other
hand, the most interesting features consists of one or few, more subtle, color charac-
teristics distributed within the image, then using the score density histograms based
on
Z
(
i.e.
, a common set of loading vectors), normalized as mentioned previously,
will be the most appropriate way to extract the desired information. These features
could consist of a vector of pixel densities corresponding to each combination of
t
i
-
t
j
values (
i
j
) obtained directly from the score histogram with a certain mesh
size, or counts of pixels falling underneath one or a few masks (see Figure 3.7).
Which type of feature to use is application dependent and often requires some trial
=
Search WWH ::
Custom Search