Figure 1.
Examples of a Rubin vase (A) and a textured figure overlying a background (B).
A: Bi-stable percept of a flower vase or two monkey faces depending on whether the borders between the luminance regions are assigned to the lighter or to the darker regions. B: The small centre square segregates from the background on basis of a difference in orientation of the line segments, and is perceived as a figure.
Figure 2.
Architecture, connectivity and input scheme of the computational model.
A: The model consists of three layers, which are unidirectional connected. Arrows define feedforward connections. The neural interactions are specific for feature preference. Lower two squares indicate the input (white regions) of the figure (left) and background (right). B: All layers receive point-to-point (retinotopic) excitatory input. Second and third layers also receive inhibitory input from all or one preceding neuron(s), respectively. C: The model input may correspond to the figure and background of a figure-ground texture as illustrated in figure 1b. Dotted lines demarcate the stimuli.
Figure 3.
Figure-ground images of filled squares (A) and frames (B).
A,B: White regions depict the input regions and black regions depict regions that provide no input to the feature specific neurons of the model. In the left column white squares represent the figures (A) and frames (B). In the right column the complementary shapes are illustrated where white regions represent the background. Dotted lines demarcate the stimuli.
Figure 4.
Figure-ground images of single squares (A) and two overlaying squares (B). Squares are shifted from the centre to illustrate one-sided border-ownership coding. Color coding is as in figure 3, except for the grey square which depicts an additional figure. Dotted lines demarcate the stimuli.
Figure 5.
Model output and firing rates of neurons located on the figural region and on the background region.
The light-dark squares in the centre column represent the NxN matrices of neurons of the model. The coloring of the matrices illustrates the membrane potential where light grey indicates high activity and dark grey zones low activity level. The white small circles depict neurons located on the figural and background regions. The arrows originating from them point to the corresponding spike responses of these neurons over time. Note that the activity pattern of the first layer of the model mirrors the texture input whereas the second layers only neurons at the figural region spike. Lower two BW squares represent the texture input, Dotted line demarcates the stimulus. Time is from stimulus onset.
Figure 6.
Distribution of spiking neurons after presenting the texture input to the model.
A: The light-dark squares in the centre column represent the NxN matrices of neurons of the model. The coloring is as in figure 5. The lines of small white circles denote an entire column of neurons of an NxN matrix. We used N-16 to clearly illustrate the distribution of the spiking pattern. T and B signify top and bottom of the matrix, respectively. Arrows point to the spiking behavior of these neurons. Dotted line demarcates the stimulus. B: The neurons from (A) are here plotted on the y-axis (small, white circles). Each black and grey dot represents a spike from the corresponding neuron on the y-axis. Spikes from neurons from feature map 1 are in black and spikes from feature map 2 neurons in grey.
Figure 7.
Illustrations of subfields of the receptive fields of layer 3 neurons.
Each neuron receives an excitatory (+) and an inhibitory (−) input from neighboring layer 2 neurons. Depending on the combination of neighbors, 8 possible distributions of subfields are possible. Grey shading indicates part of the figural region and white regions indicate the background. One-sided border ownership can be achieved when layer 3 neurons receive excitatory input from a layer 2 neuron i.e. when it is located on the figure region, and inhibitory input from a neuron located on the background. In all other cases layer 3 neurons will be silent or inhibited. The bars/arrows next to each neuron (grey circles) indicate the response.
Figure 8.
One-sided border-ownership assignment.
A: A neuron spikes when the border of the figure is on the left side of the receptive field. B: When the border is on the right side of the receptive field, the neuron does not spike. C: One-sided border-ownership for two partially overlapping figures. The neuron spikes when the border belongs to the figure at the left and not when it belongs to the right figure. Note that the receptive field stimulations are identical in both conditions. Small white circles indicate the location of the receptive field of the neuron. Dotted line demarcates the stimulus. Time is from stimulus onset.
Figure 9.
Total inhibitory input as a function of figure size.
A: The total amount of inhibitory input that a neuron receives increases for larger figures for neurons located on the figure. For neurons located on the background, the total inhibition decreases with figure size. B: Here the difference between inhibition for neurons located on the figure and for neurons on the background is plotted. Vertical dotted lines indicate the maximal figure size that the model correctly segregates.
Figure 10.
Distribution of spiking neurons after presenting four figures to the model.
A: The lines of small white circles denote an entire column of neurons of an NxN matrix. We used N-16 to clearly illustrate the distribution of the spiking of neurons. Arrows point to the spiking pattern. The light-dark squares in the centre column represent the NxN matrices of neurons of the model. The coloring is as in figure 5. Dotted line demarcates the stimulus. B: The neurons from (A) are here plotted on the y-axis (small, white circles). Each black and grey dot represents a spike from the corresponding neuron on the y-axis. Spikes from neurons from feature map 1 are in black and spikes from feature map 2 neurons in grey.
Figure 11.
Distribution of spiking neurons after presenting an outline to the model.
A: The lines of small white circles denote an entire column of neurons of the NxN matrix of the model from the different layers. We used N-16 to clearly illustrate the distribution of the spiking pattern. Arrows point to the spiking behavior of these neurons. Coding is as in figure 5. Dotted line demarcates the stimulus. B: The neurons from (A) are here plotted on the y-axis (small, white circles). Each black and grey dot represents a spike from the corresponding neuron on the y-axis. Spikes from neurons from feature map 1 are in black and spikes from feature map 2 neurons in grey.
Figure 12.
Distribution of spiking neurons after presenting four outlines to the model.
A: The lines of small white circles denote an entire column of neurons of the NxN matrix of the model from the different layers. We used N-16 to clearly illustrate the distribution of the spiking pattern. Arrows point to the spiking behavior of these neurons. Coding is as in figure 5. Dotted line demarcates the stimulus. B: The neurons from (A) are here plotted on the y-axis (small, white circles). Each black and grey dot represents a spike from the corresponding neuron on the y-axis. Spikes from neurons from feature map 1 are in black and spikes from feature map 2 neurons in grey.
Figure 13.
Responses to figure and ground as a function of stimulus contrast.
A: Average firing rate to figure and ground. B: Modulation strength (figure minus ground responses). Grey squares represent the modulation strength for different figure-ground textures observed in visual cortex of monkeys. C: Onset latencies of figure-ground modulation. Grey squares represent onset latencies for different figure-ground textures observed in monkey visual cortex. The high contrast stimulus used by Supèr (Supèr et al., 2001) is set here to 100% for comparison.
Figure 14.
Scheme illustrating the mechanisms of figure-ground segregation by the model.
A neuron located on the figure region (left panels) receives weak global inhibitory input together with retinotopic excitatory input. As a result the neuron fires spikes. In the case when a neuron receives strong global inhibition and no excitation, rebound spiking occurs. Neurons on the background region (right panels) are silent. The strong global inhibitory input cancels the excitatory drive and weak inhibitory input does not produce rebound spiking.