A Change of Border-Ownership
Increases the Backward Visual
Masking Effect
Name: Nina Vreugdenhil
Student number: 11670517
Supervisors: Doris Dijksterhuis & Matthew Self
Deadline: July 17
th2020
Abstract
In this article we investigate the effect of border-ownership on visual masking. Macknik and Livingstone (1998) suggested that the interruption of the offset response to a stimulus is an explanation for the backward masking effect. Other studies contradicted this conclusion and therefore we want to propose an alternative explanation for the masking effect. We hypothesized that a change of border-ownership increases the masking effect. We found a higher detection threshold for the condition where border-ownership changes, therefore we concluded that a change of border-ownership increases the masking effect.
1
A Change of Border-Ownership
Increases the Backward Visual Masking
Effect
N. Vreugdenhil*
*University of Amsterdam
Abstract
In this article we investigate the effect of border-ownership on visual masking. Macknik and Livingstone (1998) suggested that the interruption of the offset response to a stimulus is an explanation for the backward masking effect. Other studies contradicted this conclusion and therefore we want to propose an alternative explanation for the masking effect. We
hypothesized that a change of border-ownership increases the masking effect. We found a higher detection threshold for the condition where border-ownership changes, therefore we concluded that a change of border-ownership increases the masking effect.
1. Introduction
Visual masking is studied to get a clear view of how a percept is formed in the brain by
processing visual information. In the current study, we use a metacontrast backward masking task to study visual processing. In metacontrast backward visual masking, a mask (which does not superimpose a target, but has closely neighbouring borders) appears after onset of a target (Enns & Di Lollo, 2000). In visual masking, a target would be visible without a mask, but it can be made less visible by a mask. This depends on interactions of contours between a target and a mask, duration of a target, duration of a mask and the time between showing a target and a mask.
Processing visual stimuli relies on two channels in which neural activity is evoked. One channel evokes fast acting but short lasting responses to a visual stimulus and the other evokes slow-acting but long lasting responses to a visual stimulus (Enns & Di Lollo, 2000; Ogmen, Breitmeyer & Melvin, 2003). The fast-acting signal encodes the onset and the offset of a target. The slow-acting signal encodes the colour and the shape of a target. This two-channel theory is another explanation for the masking effect. The fast-acting response to a mask interferes with the slow-acting response to a target, which could make the target less visible.
Another explanation for the underlying mechanisms of backward visual masking based on the transient on- and offset signals, is proposed by Macknik and Livingstone (1998). They suggested that a backward mask interrupts specifically the transient offset response to a target and
therefore causes a target to be less visible. This statement suggests that the transient offset response is necessary for visibility of a target. Macknik and Livingstone tested this in humans and rhesus monkeys by using a visual masking task. They found a maximal masking effect when they stopped showing a mask 100 ms after they stopped showing a target. The time from the offset of a target to the offset of a mask is called the stimulus termination asynchrony or the STA. To study the neural correlates underlying this masking effect, they measured responses of V1 neurons in rhesus monkeys. They found that the after-discharge of the response to a target was inhibited. These two results lead to their conclusion that backward masking interrupts the transient offset response and therefore makes a target less visible.
However, the finding of Macknik and Livingstone is not the only suggested explanation for the underlying mechanisms of backward visual masking that was found. Other studies have found
2 different explanations for the visual masking effect, for example that the duration of the
response may be important for a visible target (Thompson & Schall, 1999; Kovács, Vogels & Orban, 1995; Rolls & Tovee, 1994). There are also studies that concluded that backward masking does not happen at early cortical levels, but at higher levels (Enns & Di Lollo, 2000). Which is contradicting to the finding that the offset response is responsible for visibility of the target as the offset response happens at an early level. We can conclude that the underlying mechanism of the masking effect is still unclear, therefore we will discuss an alternative explanation for the backward metacontrast masking effect.
Border-Ownership and visual masking
Border-ownership indicates to which object a border belongs and makes it possible for us to detect shapes in a correct way, therefore they help us to perceive the world around us. The distinction of shape depends on correct assignment of border-ownership (Williford & Von der Heydt, 2014). For example in figure 1, in figure 1A we see some meaningless blue regions, but when an inkblot covers these regions, like in figure 1B, we see the blue regions as letter B’s in the background. This is because some of the borders now belong to the inkblot instead of to the regions like in figure 1A.
Figure 1:Example of border-ownership. A) In this image the loose pieces own the borders. B) In this image the loose pieces turn into B’s because some of the borders undergo a change of border-ownership (From Williford, 2014).
The processing of borders is done by border-ownership cells. Zhou, Friedman and Von Der Heydt (2000) found these cells in the neural activity of monkeys. They found that these cells responded stronger when a rectangle was on the left side of a border compared to when it was on the right side of that same border (see fig. 2), or the other way around. They found this effect even when the contrast polarity in the receptive field was the same.
3 The combination of the activity of all border-ownership cells, the ones with a preference for a figure on the left and the ones with a preference for a figure on the right, will encode on which side the figure actually is (Williford & Von der Heydt, 2014). A change of border-ownership changes the strength of the activity of the cells and we suggest that this could be contributing to the masking effect.
Figure 3: The targets and masks that we used in the flipping (left) and non-flipping (right) condition are depicted in this figure. In the flipping condition the concave target owns the borders, but in the mask, the target-region is surrounded by convex objects, which will take over the borders and therefore the target-region will be seen as background. In the non-flipping condition, the convex target owns the borders and it still owns the borders when it is surrounded by concave objects in the mask. Therefore it will be seen as a figure.
To incorporate border-ownership in the backward visual masking task, we will use concave and convex objects. The Gestalt laws state that convexity strongly determines whether something is seen as a figure or as ground (Wagemans et al. 2012). Convex objects are more likely to be seen as figures than concave objects (Haushofer, Baker, Livingstone & Kanwisher, 2008). The borders will thus be assigned to convex objects if they are present. Consider the concave shaped target on the left side in figure 3. First, the target is an isolated object, therefore it “owns” the borders and it will thus be perceived as a figure on a background. If this target is surrounded by convex objects which are adjacent to the concave target, like in the mask for the flipping condition in figure 3, the borders of the target will be taken over by the convex figures according to the gestalt rules. This causes the convex objects to be perceived as figures and the concave
target-Figure 2: A border-ownership cell from the results of Zhou et al. (2000). In this figure, we can see that the measured neuron fired more often for a rectangle on the left (A & C) compared to a rectangle on the right (B & D) regardless of the colour of the rectangle and the background.
4 region to be perceived as background. The border-ownership changes, because the figure is now on the other side of the border. As shown in Zhou et al. (2000) this causes a change in the
activity of border-ownership cells that are processing these borders. We will refer to this condition as the flipping condition, because the figure ‘flips’ from one side of the border to the other side. The convex target, on the right side in figure 5, will remain owner of its borders when concave objects surround the target. Therefore it will still be perceived as a figure on a
background. There will not be a change in border-ownership in this situation and therefore the activity of the border-ownership cells will not change. We will refer to this as the non-flipping condition. We suggest that the change in neural activity caused by flipping of border-ownership increases the masking effect and therefore makes the target less visible.
The experiment
To see if a change of border-ownership indeed influences visual masking, we will compare the flipping and the non-flipping condition, by using the same stimuli that were used in the
experiment of Dijksterhuis, Roelfsema and Self (2017). They found that the masking effect was greater when border-ownership changed. We want to elaborate on this result and therefore we added more mask durations and inter-stimulus intervals or ISIs to the experiment. We want to see if the effect of border-ownership still persists. The durations are inspired by the durations that Macknik and Livingstone used.
We know that longer target durations make it easier to perceive a target as there is more time to process a target. In case of a shorter target duration, there might not be enough time to process a target, so the detection of a target might be impaired. We expect that the masking effect will be weaker for longer target durations and stronger for the shorter target durations in both the non-flipping and the non-flipping condition. However, based on the results of Dijksterhuis et al. (2017) we hypothesize that the masking effect will be greater in the flipping condition than in the non-flipping condition. Therefore we expect to see higher performance levels and a lower detection threshold for the longer target durations and lower performance levels and a higher detection threshold for the shorter target durations. Furthermore, we expect to see higher performance levels in the non-flipping condition than in the flipping condition and a higher detection threshold in the flipping condition than in the non-flipping condition. Besides, we hypothesize that the change of border-ownership will have an effect on the masking effect for all the combinations of mask durations and ISIs that we use in the experiment.
Dijksterhuis et al. (2017) found that the difference between the detection thresholds for the non-flipping condition and the non-flipping condition was greatest when the contrast of a target changed in the mask. When the contrast switched, a target was black and then became a white target- region in the mask, or the other way around. When the contrast did not switch, the target-region in the mask was the same colour as the target. To apply this finding in our experiment we will only use target and mask combinations in which the contrast switches.
2. Methods
Subjects
Participants were recruited at the Netherlands Institute for Neuroscience. In total, five participants took part in the experiment. The participants were 27 to 35 years old. The ratio man/woman was 3:2. All participants were right handed and had corrected sight.
Stimuli
The experiment was programmed and presented using Cogent in Matlab R2006b. Trials were randomized and evenly distributed. The stimuli were presented on a Trinitron T5400 monitor with a refresh rate of 85Hz. The size of the monitor was 47.6° by 39.2°. The participants were at a viewing distance of approximately 55 cm from the screen in a dark room.
5 The stimuli that we used are explained in the introduction. The height of the objects was 5.6° and the width of the objects was 3.2°. We used multiple objects that could be concave or convex. Two bars were drawn 7.7° above and below the fixation cross that was presented in the centre of the screen. A target could appear in the upper bar or in the lower bar and it could appear in the middle of that bar, 6.4° on the left of the centre of that bar or 6.4° on the right of the centre of that bar. A mask appeared on both bars and it consisted of seven convex and eight concave objects in both bars, or the other way around. The target and the other objects could be black or white.
The whole experiment contained 3024 trials. In 432 trials, no mask was presented after a target, we did this to test if a target could be detected if it was not masked. There were 1296 trials in both the masked non-flipping and the masked flipping condition.
Procedure
Figure 4: In this figure, one experimental trial is depicted. First, the participant is presented with a fixation cross, hereafter the target appears. After the target there is an interval and after the interval the mask is presented. While waiting on the response of the participant, the fixation cross is shown again. The used durations are indicated in the figure.
For this experiment we used a backward metacontrast visual masking task as described in the introduction. Before the experiment started, the participants performed a few practice trials. These trials were similar to the experimental trials. First, a fixation cross appeared for 471 ms. When the fixation cross disappeared, two horizontal bars appeared and a target was drawn in one of these bars. The participant had to indicate whether the target was in the upper bar or in the lower bar. This target was shown for 12, 24, 35, 47, 59 or 94 ms. A target could be followed by a mask or not. In the masked condition, after a target was shown there was an ISI. The ISI could be 0, 12 or 24 ms. An ISI of 0 ms meant that a target was directly followed by a mask. During the ISI, two bars without objects were visible on the screen. The bars remained the same colour as they were while the target was presented. When the interval was finished, a mask appeared. In a mask period, the two horizontal bars on the screen were filled with objects. A mask could be shown for 47 or 94 ms. The ISI and the mask duration together are called the stimulus-termination asynchrony or the STA. The STA could be 47, 59, 71, 94, 106, 118 ms. When the mask disappeared, a grey screen with a fixation cross appeared while waiting on the participants response. The participant could respond by pressing the arrow up button for the upper bar or the arrow down button for the lower bar. Hereafter, the next trial started
6 automatically. The non-masked condition was the same as the masked condition, but without an interval and a mask.
In the experiment there were four breaks. The participants could continue the experiment when they wanted by pressing a button. After the experiment the participants filled in a questionnaire that contained questions about how attentive they were during the task and how difficult the task was for them.
Analysis
The analysis of the data was done in Matlab R2019b. The psychometric function was applied to the data, using the Palamedes Toolbox. To fit the psychometric function we used a searchgrid in which the only fixed value was the guess rate, which was 0.5. From the psychometric function, the slope and the threshold per participant were obtained. The threshold indicates how difficult it was to detect a target for a certain condition. A higher threshold indicates that it was harder to detect a target. The slope is an indication of how reliable the sensory performance of a
participant is (Strasburger, 2001). We performed t-tests on the thresholds and slopes for all participants to compare the flipping and non-flipping condition. A p-value smaller than 0.05 indicated a significant difference. We also tested if there was a masking effect, this was done by comparing the masked trials to the non-masked trials. Besides, we checked if there was an effect of the location and the colour of a target. We took into account if a target was up or down and if a target was in the middle or on the side of the screen.
3. Results
We measured the proportion of correct answers given in the task. First of all, we compared the masked condition to the non-masked condition and we found that the proportion of correct answers is higher for the non-masked condition than for the masked condition (p = 0.070584). For the first three target durations, the difference between no mask and a mask is almost significantly different (p = 0.0587).
Figure 5: In this figure we can see the difference in performance levels between the no mask (blue) and the masked (red) condition averaged over all participants. On the x-axis the target durations are defined and on the y-axis the proportion of correct answers is defined.
We fitted a psychometric function on the data. From this psychometric function we derived the thresholds and slopes for the different conditions. Figure 6 is an example of the psychometric
7 function for all data of one participant. For the non-flipping (red) and the flipping (green)
condition we only used the masked trials and the non-masked condition (blue) is plotted as a baseline.
Figure 6: Psychometric function for one participant showing the proportion of correct answers for different target durations. The raw (dots) and the fitted (lines) data are plotted for the non-flipping condition (red), the flipping condition (green) and the non-masked condition (blue).
Figure 7: Detection thresholds per subject for every combination of ISI, mask duration and STA. The line in the plot is a linear line that can be used as a reference. Each point represents a combination of ISI, mask duration and STA for one subject.
In figure 7, the detection thresholds are plotted per participant and per combination of ISI, mask duration and STA. We can see that the shorter ISIs have higher thresholds and the longer ISIs have lower thresholds.
8 To see if there is a significant difference between the non-flipping and the flipping condition, we performed a t-test comparing the thresholds of the non-flipping condition and the flipping condition over all participants.
Figure 8: Thresholds for the non-flipping condition (left) and the flipping condition (right). The bars represent the average thresholds over all participants and the lines represent the thresholds per participant.
Table 1: Means and standard errors for the thresholds of the non-flipping and flipping condition and statistics derived from the t-test performed over the thresholds averaged over all participants
Mean (ms) Standard error (ms)
Non-flipping 15.7101 3.4152
Flipping 19.1123 3.9810
Non-flipping compared to flipping:
p-value t-value df SD
0.020035 -3.7449 4 2.0314
The t-test that we performed showed that the average threshold for the flipping condition (M = 19.1123, SE = 3.4152) is significantly higher than the average threshold for the non-flipping condition (M = 15.7101, SE = 3.4152) (p = 0.020035). In table 1, the statistics derived from the performed t-test are presented.
We also performed a t-test to see if there is a significant difference between the slopes for the non-flipping condition and the flipping condition
9
Figure 9: Slopes for the non-flipping condition (left) and the flipping condition (right). The bars represent the average slope over all participants and the lines represent the slopes per participant.
Table 2: Means and standard errors for the slopes of the non-flipping and flipping condition and statistics derived from the t-test performed over the slopes averaged over all participants
Mean Standard error
Non-flipping 0.1243 0.0089
Flipping 0.1027 0.0075
Non-flipping compared to flipping:
p-value t-value df SD
0.030176 3.2914 4 0.0147
The average slope for the non-flipping condition (M = 0.1243, SE = 0.0089) is significantly higher than the average slope for the flipping condition (M = 0.1027, SE = 0.0075) (p = 0.030176). The statistics derived from this t-test are presented in table 2.
We checked for other factors that could have influenced the masking effect. We took into account if the target was up or down, if the target was in the middle or on the sides and if the target was black or white. The proportion of correct answers was significantly higher when the target was in the upper bar compared to when the target was in the lower bar (p = 0.035958). The proportion of correct answers was also significantly higher when the target was black, compared to when the target was white (p = 0.042074). We compared the flipping and the non-flipping condition separately for these conditions. When the target was in the lower bar, the proportion of correct answers is significantly higher for the non-flipping condition compared to the flipping condition (p = 0.031544). When the target was white, the proportion of correct answers was significantly higher for the non-flipping condition than for the flipping condition (p = 0.044508).
10
Figure 10: The proportion of correct answers for every target duration per mask duration over the STAs, averaged over all participants. The continuous lines indicate the non-flipping condition, the dashed lines indicate the flipping condition. The different colours indicate the different target durations.
In figure 10, we can see that especially for the lower target durations, the proportion of correct answers is lower in the flipping condition than in the non-flipping condition.
11
4. Discussion
The goal of current experiment is to investigate if change in border-ownership influences the masking effect. Our results show that there was a significantly higher masking effect in the flipping condition than in the non-flipping condition. This is in line with our hypothesis and contributes to the suggestion that border-ownership plays a role in visual masking.
Figure 11: This image shows how a hypothesis about an object is formed based on border-ownership. A) For target type 1, the “figure” hypothesis wins, because the target is isolated and therefore the border-ownership cells point to the centre of the figure. In mask type 1, the convex objects will be seen as figures according to the gestalt rules. For the target-region, the “background” hypothesis will win, because the border-ownership cells point to the convex figures surrounding the target region. Therefore the target region is now seen as
background. B) In target type 2, the border-ownership cells also point to the centre of the target, because it is isolated and the “figure” hypothesis wins. In target type 2, the target region is still seen as a figure according to the gestalt rules. The border-ownership cells still point to the centre of the target region and therefore the “figure” hypothesis still wins.
A mechanism that may be underlying to our explanation of the effect of border-ownership on visual masking is recurrent processing. When a target is presented, brain areas in the visual pathway are activated one after the other, continuous higher up in the hierarchy of the visual pathway. This is called the feedforward sweep. It takes approximately 100 ms to complete this feedforward sweep. After the feedforward sweep, the brain areas that are activated stay active and there are horizontal and feedback connections that contribute to the responses in the feedforward sweep. This is called recurrent processing (Lamme & Roelfsema, 2000). In backward masking, the feedforward sweep mismatches with the horizontal and feedback
connections. A change of border-ownership may increase this mismatch which causes a stronger masking effect. Imagine a participant performing a trial of our experiment. First the participant sees an isolated concave object. The border-ownership cells project to higher visual areas and this leads to two hypotheses. One of the hypotheses states that the object is a figure and the other hypothesis states that the object is background, as shown in figure 11A. Because the target
12 is isolated, the hypothesis that states that the object is a figure is more likely and therefore the neural presentation for this hypothesis will be stronger than the neural presentation for the other hypothesis. In lower levels, the cells point to the centre of the target and the percept of the figure will be stronger, because of recurrent processing between different areas. If the target is followed by mask 1 (fig. 11A), there will be a new winning hypothesis at the higher levels. This winning hypothesis states that the previous target region is now background. The activation of border-ownership cells will change, because of the sent feedback. The cells pointing to the centre of the target region will become less active and cells pointing to the surroundings will be more active. If this change happens fast enough, there will not be enough time for the
hypothesis, that states that the object is a figure, to be confirmed. This may impair the
perception of the target. Now imagine the participant performs a trial in which target type 2 and mask type 2 are presented. Because the target is isolated, the target is, just like target type 1, seen as a figure. In mask type 2, the target region is still seen as a figure according to the gestalt rules. Therefore the feedback is in accordance with the cells that were firing to the centre of the object in lower levels and perception of the target will be less impaired.
Macknik and Livingstone found a strong masking effect for a STA of 100 ms. We did not see this effect in our results. Breitmeyer, Ro and Ogmen (2004) state that the visibility of a target in a masking task can be recovered with a second mask. In case that the visibility of the target is dependent on the offset response of a target, this seems very contradictive. If the offset would be interrupted by the mask which makes the target less visible, it would not be possible to recover the target and make it visible again by adding a second mask, because the information of the target would have already been lost.
We checked for other factors that could have influenced the masking effect like colour and location of the target. It turned out that the proportion of correct answers was lower when the target was white compared to when the target was black. The proportion of correct answers was also lower when the target was in the lower bar compared to when it was in the upper bar. Apparently, these factors made it harder to detect the target and therefore we decided to split those factors up to see if there was any effect on the effect of border-ownership. These factors did not confound our results but it might be interesting to take them into account in follow-up research.
Results from current research showed lower thresholds for longer ISI’s. This shows that the task was easier for longer ISI’s. Enns & DiLollo (1997) showed that the masking effect was greatest when the SOA was about 45 ms. The duration of the target was always 30 ms, this means that the ISI was 15 ms. After a SOA of 45 ms, the performance was increasing. This shows that the task is harder for a shorter ISI. It could also be possible that the border-ownership effect is weaker for a longer ISI. When the mask is shown right after the target, the border-ownership immediately flips in a flipping condition. When there is an interval between a target and a mask, they look like two separate images displayed after each other and the interval diminishes the effect of flipping border-ownership in visual masking.
Kovacs et al. (1995) did an experiment in which they measured the response of shape-selective neurons in a backward masking task. They found a shorter response of the shape-selective cells to the target when there was a mask, compared to when there was no mask. Also, the masking effect was greater with shorter SOAs. This means that the response duration of the shape-selective cells was too short to recognize the shape. Our findings about the effect of border-ownership may also be applicable to the shape-selective cells in the research of Kovacs et al. (1995). If the mask changes the activity of the shape selective cells, there may not be enough time to form a percept of the target, which impairs visibility of the target, just like the border-ownership cells in our research.
13 Our results show that a change of border-ownership increases the masking effect in a
metacontrast backward visual masking task and therefore it indicates that border-ownership plays a role in visual masking. In the current study, the experiment was focused on backward masking, it would be interesting to do the same experiment for forward masking. Furthermore, it would be interesting to measure neural activity from border-ownership cells during this task to investigate our theory further.
5. Reference list
Breitmeyer, B.G., Ro, T. & Ogmen, H. (2004). A comparison of masking by visual and transcranial magnetic stimulation: implications for the study of conscious and unconscious visual processing.
Consciousness and Cognition, 13, 829 – 843.
Craft, E., Schütze, H., Niebur, E. & Von der Heydt, R. (2007). A Neural Model of Figure-Ground Organization. Journal of Neurophysiology, 97(6), 4310-4326.
Dijksterhuis, D.E., Roelfsema, P.R. & Self, M.W. (2017). Backward Metacontrast Masking and the Role of Border-Ownership.
Enns, J.T. & Di Lollo, V. (1997). Object substitution: A new form of masking in unattended visual locations. Psychological Science, 8(2), 135 – 139.
Enns J.T. & Di Lollo, V. (2000). What’s New In Visual Masking? Trends in Cognitive Sciences, 4(9), 345-352.
Haushofer, J., Baker, C.I., Livingstone, M.S. & Kanwisher, N. (2008). Privileged Coding of Convex Shapes in Human Object-Selective Cortex. Journal of Neurophysiology, 100(2), 753 – 762.
Haushofer, J., Baker, C.I., Livingstone, M.S. & Kanwisher, N. (2008). Priviledged Coding of Convex Shapes in Human Object-Selective Cortex. Journal of Neurophysiology, 100(2), 753-762.
Kovács, G., Vogels, R. & Orban, G.A. (1995). Cortical Correlate of Pattern Backward Masking.
Proceedings of the National Academy of Sciences of the United States of America, 92(12),
5587-5591.
Lamme, V.A.F. & Roelfsema, P.R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neurosciences, 23(11), 571 – 579.
Macknik, S.L. & Livingstone, M.S. (1998). Neuronal Correlates of Visibility and Invisibility in the Primate Visual System. Nature Neuroscience, 1(2), 144 – 149.
Ogmen, H., Breitmeyer, B.G. & Melvin, R. (2003). The What and Where in Visual Masking. Vision
Research, 43(12), 1337 – 1350.
Reynolds, J.H., Pasternak, T. & Desimone, R. (2000). Attention Increases Sensitivity of V4 Neurons. Neuron, 26, 703 – 714.
Rolls, E.T. & Tovee, M.J. (1994). Processing Speed in the Cerebral Cortex and the Neurophysiology of Visual Masking. Proceedings. Biological Sciences, 257(1348), 9-15.
Strasburger, H. (2001). Converting Between Measures of Slope of the Psychometric Function.
Perception & Psychophysics, 63(8), 1348 – 1355.
Thompson, K. G., & Schall, J. D. (1999). The Detection of Visual Signals by Macaque Frontal Eye Field During Masking. Nature Neuroscience, 2(3), 283–288.
14 Wagemans, J., Elder, J.H., Kubovy, M., Palmer, S.E., Peterson, M.A., Singh, M. & Von der Heydt, R. (2012). A Century of Gestalt Psychology in Visual Perception I. Perceptual Grouping and Figure-Ground Organization. Psychological Bulletin, 138(6), 1172-1217.
Williford, J.R. (2014). Neural Basis of Perceptual Organization of Natural Scenes: Emergence of Object-Based Coding in the Primate Visual Cortex.
Williford, J.R. & Von der Heydt, R. (2014). Border-ownership Coding. Scholarpedia Journal, 8(10). Zhou, H., Friedman, H.S. & Von der Heydt, R. (2000). Coding of Border Ownership in Monkey Visual Cortex. The Journal of Neuroscience, 20(17), 6594 – 6611.
