Environmental Engineering Reference
In-Depth Information
slice of
X
along λ is filled with such numbers. Linear combinations of these num-
bers for the red, green, and blue channels will give rise to any colors of the visible
spectra.
In summary, a multivariate digital image is a three-way array of data
X
from
which it is possible to compute or extract various features, such as geometri-
cal/morphological, colors, and textural features. Some methods and algorithms used
for this purpose will be presented in the next section.
3.4 Machine Vision Framework and Methods
As mentioned earlier in the introduction section, the information to be extracted
from multivariate images of processes and products are essentially stochastic, are
generally more complex, and often require combining different methods for extract-
ing relevant image features (
i.e.
color, textural features,
etc.
) and for correlating them
to key process variables. A general machine vision framework proposed by Liu [13]
and Tessier
et al.
[26] is presented in Figure 3.4. This three-step framework will be
used to discuss the role of each multivariate method and how they interconnect in
order to obtain valuable information from process images.
Image acquisition
Pre-processing
{
─
MIA
─
MRA
─
MR-MIA
─
Others…
─
PCA
─
Feature extraction
{
PLS
─
PLS-DA
─
Others…
Feature reduction / analysis
Desired information
Figure 3.4
A general machine vision framework used in multivariate imaging. Typical multivariate
methods used in each step are also listed. Adapted from Liu [13]
The steps to be performed following image acquisition are described below:
•
Step 1: Image pre-processing.
When necessary, image quality is enhanced to
improve feature extraction. Examples include contrast and brightness adjust-
ments, correction for non-uniform illumination, noise removal, edge enhance-
ment and segmentation. These operations are performed using well known tra-
Search WWH ::
Custom Search