Rockwell Automation 5370-CVIM2 Module - Manual (Page 330)

Chapter 7
Inspection Tools
7–92
From the Image Kernel panel you can select a spatial filtering function
using a 3x3 or 5x5 kernel with either fixed or user–configurable coefficients,
to perform neighborhood operations on the image tool pixels. The basic
purpose of these operations is to either enhance the gradients that represent
edges on image features, or to filter noise from the image.
Here is a brief description of each kernel in the Image Kernel panel:
•
Sobel X –– This is a 3x3 directional kernel that enhances the vertical
gradients on an inspected object.
•
Sobel Y –– This is a 3x3 directional kernel that enhances the horizontal
gradients on an inspected object.
•
Laplace –– This is a 3x3 non–directional kernel that enhances vertical
and horizontal gradients on an inspected object.
•
X Edge –– This is a 5x5 directional kernel that enhances the vertical
gradients on an inspected object.
•
Y Edge –– This is a 5x5 directional kernel that enhances the horizontal
gradients on an inspected object.
•
XY Edge –– This is a 5x5 non–directional kernel that enhances the
vertical and horizontal gradients on an inspected object.
•
Average 3x3 –– This is a 3x3 kernel that performs an averaging function
on each pixel in the image.
•
Average 5x5 –– This is a 5x5 kernel that performs an averaging function
on each pixel in the image.
•
User 3x3 –– This is a 3x3 kernel in which the coefficient values are
user–selectable.
•
User 5x5 –– This is a 5x5 kernel in which the coefficient values are
user–selectable.
The kernel performs its neighborhood operation identically for all spatial
filter functions. That is, the kernel “scans” the image tool left–to–right and
top–to–bottom, and each coefficient in the 9 (or 25) matrix multiplies the
gray scale value of the corresponding pixel in the underlying 3x3 (or 5x5)
part of the image tool. These 9 (or 25) multiplications are then summed, and
the total is placed in the image tool pixel directly under the center of the
matrix. This process is repeated for each pixel in the image tool.
NOTE: All kernels except the “average” kernels produced signed images;
that is, they may have negative values. The “average” kernels produce only
unsigned images –– their values are always positive.
Figure 7.76 (page 7–93) provides an example that illustrates the
neighborhood operation of a kernel using, for this purpose, the coefficient
array and values of a Sobel X filter. The example shows the result of the
kernel’s neighborhood operations across the second row of pixels from the
top of the image tool.