Centre-Surround cell (DoG) operator and Dot-pattern selective cell operator

N. Petkov and W.T. Visser, University of Groningen,
Department of Computing Science, Intelligent Systems

This page contains examples concerning a simulation program available on internet.


Examples


1) Effect of surround inhibition on excitatory centre responses

Select the following parameter values:
  • Input image
synthetic1.png
  • Output image
Centre-surround cell
  • Map circles onto result
off
  • Invert output
off
  • Spot radii
2.0000, 5.5000, 9.5000, 12.0000, 24.0000
  • Number of different radii
1
  • Centre-surround size ratio
0.5
  • Normalisation weight
1
  • Channels
ON channel

The additional parameters do not influence the result of the method and can ignored.

Observe that, by selecting a single radius for display, the small spot stimuli in the upper left of the image invoke a high response when a spot radius of 2.0000 is selected for display. Larger radii result in a much more wide, although less high response. Large spot stimuli invoke high centre-surround responses when e.g. a spot radius of 24.0000 is selected for display (although varying with the stimulus radius). When radii significantly smaller than the stimulus radius are selected, only the spot contour survives surround inhibition.


2) Properties of spot detection results

2.1) Low responses of near-image boundary stimuli

Select the following parameter values:
  • Input image
diffsize-inp.png
  • Output image
Spot-detecting subunit
  • Map circles onto result
off
  • Invert output
off
  • Spot radii
4.0000
  • Number of different radii
1
  • Centre-surround size ratio
0.5
  • Normalisation weight
1
  • Channels
ON channel
  • Response ratio threshold
0.8
  • Inhibitory elements
20

The additional parameters do not influence the result of the method and can ignored.

Observe the detected spots positioned on the edge of the image, which yield a low response with respect to more centrally located other detected spots. It is induced by the lack of stimuli outside the visual field (observe the centre-surround cell output using the same parameter values). This effect can be found in human vision as well when stimuli at the farmost eye angle: a blurred response results with lower probability of spot detection with respect to stimuli presented in the centre of the visual field.

2.2) Multi-radii processing

Most parameters have values as in example 2.1, except for the spot radii.
  • Input image
diffsize-inp.png
  • Output image
Spot-detecting subunit
  • Map circles onto result
off
  • Invert output
off
  • Spot radii
3.7000, 4.3000
  • Number of different radii
1
  • Centre-surround size ratio
0.5
  • Normalisation weight
1
  • Channels
ON channel
  • Response ratio threshold
0.8
  • Inhibitory elements
20

The additional parameters do not influence the result of the method and can ignored.

Successively select spot radii 3.8000 and 4.3000 to be displayed. The detection of a uniformly sized pattern in example 2.1 has now been split up into two differently sized spots. Although many spots with a radius of 3.8000 have been detected, most of them can be found on the outer sides of the image. This is again caused by a lack of spot stimuli around these spots (outside the image boundary), which can be observed by inspecting the centre-surround cell output using the same parameter configuration.


3) Effect of ON, OFF and combined channel responses

Select the following parameter values:
  • Input image
dualchannel.png
  • Output image
Spot-detecting unit
  • Map circles onto result
off
  • Invert output
off
  • Spot radii
3.0000, 11.0000, 20.0000 26.0000 50.0000
  • Number of different radii
1
  • Centre-surround size ratio
0.5
  • Normalisation weight
1
  • Channels
< variable >
  • Response ratio threshold
0.5
  • Inhibitory elements
20

The additional parameters do not influence the result of the method and can ignored.

Initially, select the ON channel to be processed. Five differently sized spots will be detected corresponding to the spot radii parameter list. When one selects the OFF channel the previously found spots will no longer be detected as the method has now become sensitive to dark spots on a light background only. By selecting ON and OFF channel both spot types will be detected.


4) Difference in binary and density spot-group response

Select the following parameter values:
  • Input image
checkerboard.png
  • Output image
Spot-group detecting unit
  • Map circles onto result
off
  • Invert output
off
  • Spot radii
5.5000
  • Number of different radii
1
  • Centre-surround size ratio
0.5
  • Normalisation weight
1
  • Channels
ON channel
  • Response ratio threshold
0.8
  • Inhibitory elements
20
  • Spot response threshold
0.5
  • Window size
2
  • Minimum number of spots
0
  • Unit output
< variable >

The additional parameters do not influence the result of the method and can ignored.

Observe the spot-group detecting unit response using binaryoutput. The small black lines, which consist of pixel not considered element of a uniformly sized spot group, are caused by limitations of the spot-group detecting unit parameter configuration. A density output produces the same non-group exceptions but shows some more detail about the density of the detected groups. By selecting this configuration the output contains, besides a constant response, some more strong responses. These responses are caused tiny, one pixel sized shifts in the spot-detecting subunit relative to other responses in the output.


5) Results of various smoothing sizes in the spot-density averaging unit

Select the following parameter values:
  • Input image
detect_inp.png
  • Output image
Spot-density averaging unit
  • Map circles onto result
off
  • Invert output
off
  • Spot radii
2.6500
  • Number of different radii
1
  • Centre-surround size ratio
0.5
  • Normalisation weight
1
  • Channels
ON channel
  • Response ratio threshold
0.8
  • Inhibitory elements
20
  • Spot response threshold
0.5
  • Window size
2
  • Minimum number of spots
0
  • Unit output
density
  • Neighbourhood
< variable >

The additional parameters do not influence the result of the method and can ignored.

Observe the response of the spot-density averaging unit using a neighbourhood of 1. The produced result shows to contain inconsistencies in its response: separating low-response lines are still visible. The response consistency can be improved by enlarging the neighbourhood. A parameter value of 5 already produces good results.


Last changed: 2008-07-03