Example Operator Networks
To illustrate the tableau capabilities and also show some standard image processing
sequences, we provide some example networks:
- Adaptive thresholding example in file athdemo.hjv:
This demo compares the adaptive thresholding operator with
an attempt to normalize by the illumination field and then
threshold. Which is more successful? Does changing the
amount of smoothing or type of smoothing operator have an effect?
- Change detection example in file chgdemo.hjv:
This example shows change detection when the illumination
and scene are largely constant. What is the distribution
of pixel difference values?
- Region segmentation and labeling example in file clsdemo.hjv:
Compares the classifacation operator with explicit selective
thresholding and merging. Could you also isolate the stripe
on the pen as a separate class?
- Edge detector comparison example in file edgdemo.hjv:
Compares the region boundary from the Roberts Cross and
Canny edge detectors with the boundary tracked around the
region found by thresholding. Can you choose a threshold
that will allow all of the object region to be isolated?
- Intensity manipulation example in file eqdemo.hjv:
This example compares the effects of histogram equalization
and contrast normalization. Look at the histograms to understand
the differences. Which approach is more effective here and why?
- Another intensity rescaling example in file expdemo.hjv:
This example shows effects of the logarithmic, exponential and
raise to a power operators. Considerable rescaling was needed
to get images where the white regions had similar intensities.
Look at their histograms to understand
why the result images are very different.
- FFT example in file fftdemo.hjv:
This illustrates how to extract the regular vertical stripes in the image background.
- Noise removal example in file fltdemo.hjv:
Showing the effect of several different filtering algorithms
on an image corrupted with speckle noise. Which is better for this
particular image and noise? What effect does changing the kernel
shape have? Would you get better results with applying median
smoothing after conservative smoothing?
- Geometric transformation example in file geodemo.hjv:
This examples shows how an affine transformation can be use to
implement a rotation, translation, scale and reflection.
The image dimensions and clipping are a consequence of the
implementation of the individual operators, and are not
fundamental to the operation of the operators. Also, the
individual rotation operator rotates about the center of the
image, whereas the affine operator rotates about the image
origin, which is at the upper left. This means that a different
translation value is needed to align the results, but this is
not a difference in principle.
- Hough transform example in file hghdemo.hjv:
Overlaying the detected lines over the original image, also
using boundary detection on a binary image. What happens if
you change the Hough detection parameter to get the missing
- Hit and Miss operator example in file htmdemo.hjv:
This compares the Hit and Miss operator and an alternative
approach using the thinning operator. The Hit and Miss operator
output is thickened.
- Line detection example in file lnddemo.hjv:
This compares two different approaches to detecting vertical
light lines against a dark background. Is there a difference?
- Image logic example in file logdemo.hjv:
This network shows a number of different logical operators
- Binary image morphology example in file mordemo.hjv:
This example compares the effects of the standard four
image morphology operators.
- Bitshift or rescale example in file shfdemo.hjv:
This compares two different approaches to dividing image
intensities by 2. Is there a difference?
- Skeleton example in file skldemo.hjv:
Illustrates the relationship between the distance transform,
the medial axis transform and the skeleton.
- Edge thinning example in file thndemo.hjv:
Compares the thinning possible from a given kernel on
the output of the Sobel and Compass edge detectors.
Can you add a new kernel for thinning the horizontal-ish
- Unsharp filtering example in file unsdemo.hjv:
This computes the unsharp filter and also the explicit
approach with gaussian smoothing and image arithmetic.
A histogram shows a few pixels differ significantly and
thresholding shows where these pixels are. What is the
- Zero crossing example in file zcdemo.hjv:
Compares the Laplacian of Gaussian, the Laplacian on
a hand-smoothed image, and the zero crossings found
on the Laplacian output. How well do the operators
compare on more complex images?
When loading these demonstration networks, you do not need to specify a
full URL, and instead only enter the associated filename.
©2003 R. Fisher, S. Perkins,
A. Walker and E. Wolfart.