Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Flexible gating of contextual influences in natural vision

Abstract

Identical sensory inputs can be perceived as markedly different when embedded in distinct contexts. Neural responses to simple stimuli are also modulated by context, but the contribution of this modulation to the processing of natural sensory input is unclear. We measured surround suppression, a quintessential contextual influence, in macaque primary visual cortex with natural images. We found that suppression strength varied substantially for different images. This variability was not well explained by existing descriptions of surround suppression, but it was predicted by Bayesian inference about statistical dependencies in images. In this framework, surround suppression was flexible: it was recruited when the image was inferred to contain redundancies and substantially reduced in strength otherwise. Thus, our results reveal a gating of a basic, widespread cortical computation by inference about the statistics of natural input.

This is a preview of subscription content, access via your institution

Access options

Rent or buy this article

Prices vary by article type

from$1.95

to$39.95

Prices may be subject to local taxes which are calculated during checkout

Figure 1: Variability of surround modulation with natural images.
Figure 2: Standard and flexible normalization models of surround suppression.
Figure 3: Drive to the surround does not explain surround suppression strength.
Figure 4: Surround divisive normalization is optimal only for statistically homogeneous stimuli.
Figure 5: Standard and flexible normalization differ most for balanced image ensembles.
Figure 6: Surround suppression strength depends on image homogeneity.
Figure 7: Homogeneity depends on neuronal tuning.

Similar content being viewed by others

References

  1. Schwartz, O., Hsu, A. & Dayan, P. Space and time in visual context. Nat. Rev. Neurosci. 8, 522–535 (2007).

    Article  PubMed  Google Scholar 

  2. Kohn, A. Visual adaptation: physiology, mechanisms and functional benefits. J. Neurophysiol. 97, 3155–3164 (2007).

    Article  PubMed  Google Scholar 

  3. Carandini, M. & Heeger, D.J. Normalization as a canonical neural computation. Nat. Rev. Neurosci. 13, 51–62 (2012).

    Article  CAS  Google Scholar 

  4. Li, Z. Contextual influences in V1 as a basis for pop out and asymmetry in visual search. Proc. Natl. Acad. Sci. USA 96, 10530–10535 (1999).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Itti, L. & Koch, C. Computational modelling of visual attention. Nat. Rev. Neurosci. 2, 194–203 (2001).

    Article  CAS  PubMed  Google Scholar 

  6. Clifford, C.W.G., Wenderoth, P. & Spehar, B. A functional angle on some after-effects in cortical vision. Proc. Biol. Sci. 267, 1705–1710 (2000).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Ohshiro, T., Angelaki, D.E. & DeAngelis, G.C. A normalization model of multisensory integration. Nat. Neurosci. 14, 775–782 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  8. Louie, K., Khaw, M.W. & Glimcher, P.W. Normalization is a general neural mechanism for context-dependent decision making. Proc. Natl. Acad. Sci. USA 110, 6139–6144 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Gilbert, C.D. & Wiesel, T.N. The influence of contextual stimuli on the orientation selectivity of cells in primary visual cortex of the cat. Vision Res. 30, 1689–1701 (1990).

    Article  CAS  PubMed  Google Scholar 

  10. Knierim, J.J. & van Essen, D.C. Neuronal responses to static texture patterns in area V1 of the alert macaque monkey. J. Neurophysiol. 67, 961–980 (1992).

    Article  CAS  PubMed  Google Scholar 

  11. Angelucci, A. & Bressloff, P.C. Contribution of feedforward, lateral and feedback connections to the classical receptive field center and extra-classical receptive field surround of primate V1 neurons. Prog. Brain Res. 154, 93–120 (2006).

    Article  PubMed  Google Scholar 

  12. Sceniak, M.P., Ringach, D.L., Hawken, M.J. & Shapley, R. Contrast's effect on spatial summation by macaque V1 neurons. Nat. Neurosci. 2, 733–739 (1999).

    Article  CAS  PubMed  Google Scholar 

  13. Cavanaugh, J.R., Bair, W. & Movshon, J.A. Nature and interaction of signals from the receptive field center and surround in macaque V1 neurons. J. Neurophysiol. 88, 2530–2546 (2002).

    Article  PubMed  Google Scholar 

  14. Levitt, J.B. & Lund, J.S. Contrast dependence of contextual effects in primate visual cortex. Nature 387, 73–76 (1997).

    Article  CAS  PubMed  Google Scholar 

  15. Walker, G.A., Ohzawa, I. & Freeman, R.D. Asymmetric suppression outside the classical receptive field of the visual cortex. J. Neurosci. 19, 10536–10553 (1999).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  16. Cavanaugh, J.R., Bair, W. & Movshon, J.A. Selectivity and spatial distribution of signals from the receptive field surround in macaque V1 neurons. J. Neurophysiol. 88, 2547–2556 (2002).

    Article  PubMed  Google Scholar 

  17. Sillito, A.M., Grieve, K.L., Jones, H.E., Cudeiro, J. & Davis, J. Visual cortical mechanisms detecting focal orientation discontinuities. Nature 378, 492–496 (1995).

    Article  CAS  PubMed  Google Scholar 

  18. Vinje, W.E. & Gallant, J.L. Sparse coding and decorrelation in primary visual cortex during natural vision. Science 287, 1273–1276 (2000).

    Article  CAS  PubMed  Google Scholar 

  19. Haider, B. et al. Synaptic and network mechanisms of sparse and reliable visual cortical activity during nonclassical receptive field stimulation. Neuron 65, 107–121 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Ozeki, H., Finn, I.M., Schaffer, E.S., Miller, K.D. & Ferster, D. Inhibitory stabilization of the cortical network underlies visual surround suppression. Neuron 62, 578–592 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  21. Carandini, M. et al. Do we know what the early visual system does? J. Neurosci. 25, 10577–10597 (2005).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  22. Olshausen, B.A. & Field, D.J. Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature 381, 607–609 (1996).

    Article  CAS  PubMed  Google Scholar 

  23. Bell, A.J. & Sejnowski, T.J. The “independent components” of natural scenes are edge filters. Vision Res. 37, 3327–3338 (1997).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  24. Schwartz, O. & Simoncelli, E.P. Natural signal statistics and sensory gain control. Nat. Neurosci. 4, 819–825 (2001).

    Article  CAS  PubMed  Google Scholar 

  25. Karklin, Y. & Lewicki, M.S. Emergence of complex cell properties by learning to generalize in natural scenes. Nature 457, 83–86 (2009).

    Article  CAS  PubMed  Google Scholar 

  26. Spratling, M.W. Predictive coding as a model of response properties in cortical area V1. J. Neurosci. 30, 3531–3543 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Coen-Cagli, R., Dayan, P. & Schwartz, O. Cortical surround interactions and perceptual salience via natural scene statistics. PLOS Comput. Biol. 8, e1002405 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Lochmann, T., Ernst, U.A. & Deneve, S. Perceptual inference predicts contextual modulations of sensory responses. J. Neurosci. 32, 4179–4195 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  29. Vinje, W.E. & Gallant, J.L. Natural Stimulation of the nonclassical receptive field increases information transmission efficiency in V1. J. Neurosci. 22, 2904–2915 (2002).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  30. Webb, B.S., Dhruv, N.T., Solomon, S.G., Tailby, C. & Lennie, P. Early and late mechanisms of surround suppression in striate cortex of macaque. J. Neurosci. 25, 11666–11675 (2005).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  31. Heeger, D.J. Normalization of cell responses in cat striate cortex. Vis. Neurosci. 9, 181–197 (1992).

    Article  CAS  PubMed  Google Scholar 

  32. Barlow, H.B. Possible principles underlying the transformations of sensory messages. in Sensory Communication (ed. Rosenblith, W.A.) 217–234 (MIT Press, 1961).

  33. Ruderman, D.L. & Bialek, W. Statistics of natural images: scaling in the woods. Phys. Rev. Lett. 73, 814–817 (1994).

    Article  CAS  PubMed  Google Scholar 

  34. Felsen, G., Touryan, J., Han, F. & Dan, Y. Cortical sensitivity to visual features in natural scenes. PLoS Biol. 3, e342 (2005).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  35. Pouget, A., Beck, J.M., Ma, W.J. & Latham, P.E. Probabilistic brains: knowns and unknowns. Nat. Neurosci. 16, 1170–1178 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Nauhaus, I., Busse, L., Carandini, M. & Ringach, D.L. Stimulus contrast modulates functional connectivity in visual cortex. Nat. Neurosci. 12, 70–76 (2009).

    Article  CAS  PubMed  Google Scholar 

  37. Rubin, D.B., Van Hooser, S.D. & Miller, K.D. The stabilized supralinear network: a unifying circuit motif underlying multi-input integration in sensory cortex. Neuron 85, 402–417 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Ahmadian, Y., Rubin, D.B. & Miller, K.D. Analysis of the stabilized supralinear network. Neural Comput. 25, 1994–2037 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  39. Adesnik, H., Bruns, W., Taniguchi, H., Huang, Z.J. & Scanziani, M. A neural circuit for spatial summation in visual cortex. Nature 490, 226–231 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Nienborg, H. et al. Contrast dependence and differential contributions from somatostatin- and parvalbumin-expressing neurons to spatial integration in mouse V1. J. Neurosci. 33, 11145–11154 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Pfeffer, C.K., Xue, M., He, M., Huang, Z.J. & Scanziani, M. Inhibition of inhibition in visual cortex: the logic of connections between molecularly distinct interneurons. Nat. Neurosci. 16, 1068–1076 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  42. Lee, S., Kruglikov, I., Huang, Z.J., Fishell, G. & Rudy, B. A disinhibitory circuit mediates motor integration in the somatosensory cortex. Nat. Neurosci. 16, 1662–1670 (2013).

    CAS  PubMed  PubMed Central  Google Scholar 

  43. Pi, H.-J. et al. Cortical interneurons that specialize in disinhibitory control. Nature 503, 521–524 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. Schwartz, O., Sejnowski, T.J. & Dayan, P. Soft mixer assignment in a hierarchical generative model of natural scene statistics. Neural Comput. 18, 2680–2718 (2006).

    Article  PubMed  PubMed Central  Google Scholar 

  45. Yamins, D.L.K. et al. Performance-optimized hierarchical models predict neural responses in higher visual cortex. Proc. Natl. Acad. Sci. USA 111, 8619–8624 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Gilbert, C.D. & Li, W. Top-down influences on visual processing. Nat. Rev. Neurosci. 14, 350–363 (2013).

    Article  CAS  PubMed  Google Scholar 

  47. Kersten, D., Mamassian, P. & Yuille, A. Object perception as Bayesian inference. Annu. Rev. Psychol. 55, 271–304 (2004).

    Article  PubMed  Google Scholar 

  48. Gershman, S.J. & Niv, Y. Learning latent structure: carving nature at its joints. Curr. Opin. Neurobiol. 20, 251–256 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Green, C.S., Benson, C., Kersten, D. & Schrater, P. Alterations in choice behavior by manipulations of world model. Proc. Natl. Acad. Sci. USA 107, 16401–16406 (2010).

    Article  PubMed  PubMed Central  Google Scholar 

  50. Jia, X., Smith, M.A. & Kohn, A. Stimulus selectivity and spatial coherence of gamma components of the local field potential. J. Neurosci. 31, 9390–9403 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  51. Simoncelli, E.P., Freeman, W.T., Adelson, E.H. & Heeger, D.J. Shiftable multiscale transforms. IEEE Trans. Inf. Theory 38, 587–607 (1992).

    Article  Google Scholar 

  52. Carandini, M., Heeger, D.J. & Movshon, J.A. Linearity and normalization in simple cells of the macaque primary visual cortex. J. Neurosci. 17, 8621–8644 (1997).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  53. Portilla, J. & Simoncelli, E. A Parametric texture model based on joint statistics of complex wavelet coefficients. Int. J. Comput. Vis. 40, 49–70 (2000).

    Article  Google Scholar 

  54. Stocker, A.A. & Simoncelli, E.P. Noise characteristics and prior expectations in human visual speed perception. Nat. Neurosci. 9, 578–585 (2006).

    Article  CAS  PubMed  Google Scholar 

Download references

Acknowledgements

We thank P. Dayan, C.A. Henry and A. Huk for comments on an earlier version of this manuscript, members of the Kohn laboratory for help performing recordings, and S. Barthelme for discussion on estimating model performance. This work was supported by a US National Institutes of Health grant to O.S. and A.K. (CRCNS EY021371), an Irma T. Hirchl Career Scientist Award (A.K.), a Sloan Research Fellowship (O.S.) and by Research to Prevent Blindness.

Author information

Authors and Affiliations

Authors

Contributions

R.C.-C., A.K. and O.S. designed the study. R.C.-C. collected and analyzed the data. R.C.-C., A.K. and O.S. wrote the manuscript.

Corresponding author

Correspondence to Ruben Coen-Cagli.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Integrated supplementary information

Supplementary Figure 1 Measures of surround energy.

(a) Cross-validated prediction quality, averaged across neurons, for the standard (orange) and flexible (green) models. Each pair of whisker plots corresponds to a distinct constellation of surround filters or instantiation of the normalization model, as explained below. Boxes denote the 25–75th percentile; whiskers, 10–90th percentile; white line, median; white circle, mean. In all cases, the flexible model out-performed the standard model (p<0.0001). We note that although there was little variation in average performance across different instantiations of surround tuning, we did find that some instantiations performed better than others in each unit. The difference in prediction quality between the best and worst measure was larger than 25% in 63/126 neurons, and larger than 50% in 34/126 of the neurons. However, no definition of the surround consistently performed better than any another. (b) Average MR across all neurons, for images providing weak or strong surround drive (as in main Figure 3b). Conventions as in (a), except white circle indicates geometric mean. In most cases, suppression was not stronger when surround drive was strong vs. weak (p ≥ 0.02 in all cases, except “image-based”, “Lp-norm” and “Free exp” for which p<0.0001). To test the robustness of our results to other measures of surround drive, we considered the following variations on the model presented in the main manuscript: (1) Surround strongly tuned for orientation and spatial frequency (denoted “narrow ori, sf”): The surround was composed of filters matched to those representing the RF, including the same orientation preference (which was fixed across images). In total, this surround used 16 filters, 8 locations x 2 for the quadrature pair at each location. (2) Surround strongly tuned for orientation but not for spatial frequency (denoted “narrow ori, broad sf”): The surround was composed of filters matched to those representing the RF, but with 3 different spatial frequency preferences. In total, this surround used 48 filters, 8 locations x 3 spatial frequencies at each location x 2 for the quadrature pair at each location and spatial frequency. (3) Surround tuned for spatial frequency but not for orientation (denoted “broad ori, narrow sf”): The surround was composed of filters matched to those representing the RF, but with 4 different orientation preferences. In total, this surround used 64 filters, 8 locations x 4 orientations at each location x 2 for the quadrature pair at each location and orientation. (4) Surround broadly tuned for orientation and spatial frequency (denoted “broad ori, sf”): The surround was composed of filters with 4 different orientations and 3 different spatial frequencies. In total, this surround used 192 filters, 8 locations x 4 orientations at each location x 3 spatial frequencies at each location x 2 for the quadrature pair at each location, orientation, and spatial frequency. (5) Spatially in-homogeneous surround (denoted “location-based”): We also considered the possibility that surround suppression was spatially inhomogeneous (Cavanaugh et al. J Neurophysiol 88, 2547-2556 (2002); Walker et al. J Neurosci 19, 10536-10553 (1999)). We used the filters in configuration (3) above, but allowed the gain to vary across spatial locations. The two locations at the ends of the RF (collinear with the RF-preferred orientation) had a gain γcol; the two locations at the sides (parallel but not collinear) of the RF had a gain γpar; and the two pairs of locations along the diagonals had gains γdiag1, γdiag2. Thus, the surround activity for a given image, γEs, is replaced by (γcolEs-col + γparEs-par + γdiag1Es-diag1 + γdiag2Es-diag2). (6) Normalization with Lp norm: In the main text we considered a form of the normalization model used previously for surround suppression (Cavanaugh et al. J Neurophysiol 88, 2530-2546 (2002)). We chose this form because it has a minimal number of free parameters. With this choice, the normalization pool computes the L2 norm of the filters (square root of the sum of squared filter outputs, which amounts to image contrast in the filters’ frequency and orientation band) followed by a fixed exponent of 1. In a more general form of the normalization model (Carandini and Heeger Nat Rev Neurosci 13, 51-62 (2011)), the normalization pool computes an Lp norm (p-th root of the sum of p-th power of filters outputs, with p a free parameter rather than fixed to 2), and the numerator and denominator can use different exponents. To be sure that our results did not rely on the particular way in which combined filter outputs, we fit two additional models using the image-based surround used in the main text: (i) we calculated the drive to the RF and surround using an Lp norm, where p is a free parameter, rather than the L2 norm (denoted “Lp-norm”); (ii) we used the L2 norm but allowed the exponent for the numerator and denominator to be different from each other and different from 1 (denoted “Free exp.”). Finally, in order to provide a baseline against which to compare the standard normalization models, we also estimated prediction quality for a ‘center model’ without normalization. Responses in this model are defined by the numerator of the standard normalization model (eq. (1) of Methods). For this model, the prediction quality was on average 0.13 (c.i.=[0.06 0.21]), only slightly better than the null model which predicts that the response to each image is equal to the average response across images.

Supplementary Figure 2 Continuous gating of surround suppression.

(a,b) Distribution of inferred probabilities of homogeneity (abscissa), for the images used in the experiment (a) and a larger database of natural images (N=10,000) used to train the Bayesian model (b). Each image contributes multiple times, with different values depending on the filters used for the inference. Inferred probabilities were usually near zero or 1. We verified that a flexible model in which we used the inferred probability for gating (rather than discretizing the probability to a 0 or 1) did not improve the predictions, presumably because most images had probabilities near 0 or 1. (c) To test whether the gating of the surround was continuous or binary, we conducted an additional experiment, recording 91 neurons in 1 animal using methods identical to those in the main text. We used a subset of the experimental images, which consisted of the 10 most frequently homogeneous images and the 10 most frequently heterogeneous images. We used two manipulations to vary the probabilities in a continuous way. First, we rotated the surround of the images by 15, 30 and 90 degrees with respect to the center, reducing the probability that the image is inferred to be homogeneous. Second, we morphed pairs of images, in which one image had an inferred probability close to 0 and the other close to 1, by taking linear combinations of the images with relative weights of 0.25/0.75, 0.50/0.50, and 0.75/0.25, leading to intermediate values of inferred probability of homogeneity. This panel shows the average NMR (defined in main text Methods) across images and neurons with different levels of inferred probability (abscissa); solid green line, best linear fit (weighted least squares). Suppression strength varies linearly with inferred probability of homogeneity, suggesting that the gating of the surround is continuous, not binary. (d) Black symbols, average surround drive (defined in main text Methods) across images and neurons with different levels of inferred probability (abscissa); solid green line, best linear fit (weighted least squares). There is little dependence of drive on inferred homogeneity, suggesting that the trends in (c) cannot be explained by changes in surround drive. Error bars and dashed lines represent 68% c.i.

Supplementary Figure 3 Comparison of standard and flexible model predictions.

(a) Predicted MR vs. observed MR for two example neurons (rows). Values on the abscissa are repeated in the two columns, to compare the predictions of the standard and flexible models for the same neuron. Blue symbols are for images classified as homogeneous by the Bayesian model; red symbols, heterogeneous. Error bars denote 68% c.i. (b) Predicted MR vs. observed MR across all images and neurons. Dots represent means, error bars denote 68% c.i. In both the example neurons and the population, the standard model tends to predict stronger suppression than observed (smaller MR values), particularly for heterogeneous images (red symbols in (a)).To be sure that the improved prediction quality of the flexible model was not due simply to muting the surround in a random subset of images, we also verified that a flexible model with random inference about image homogeneity was not better than the standard normalization model (prediction quality 0.5, c.i. [0.42 0.56]; p=0.41 for the comparison to the standard model).

Supplementary Figure 4 Inference about homogeneity measures center-surround similarity.

The map depicts the probability that visual inputs producing center and surround energies (defined in Supplementary Modeling 2), respectively, are inferred homogeneous by the MGSM model. The thick, black contours separate regions with probability larger vs. smaller than 0.5. (b) Same as (a), but computed using the measure of similarity between center and surround Pk, Ps defined in Supplementary Modeling 2 (α = 7; β = 1.7α). The heuristic (b) provides a reasonable approximation to the full Bayesian inference (a).

Supplementary Figure 5 Reduced surround suppression with images that lack natural structure.

(a) To test the prediction that images lacking higher-order statistical dependencies are heterogeneous, and therefore should not be suppressive, we conducted an additional experiment. We recorded 91 neurons in 1 animal, using methods identical to those in the main text. We created white noise images—which lack statistical dependencies—by choosing the luminance value of each pixel from a Gaussian distribution, with mean set to the mean screen luminance, and standard deviation set to 25% of the pixel range [0 255]. Samples outside that range were clipped. Each pixel was 0.08 x 0.08 degrees in size. We created 10 large (6.7 degrees diameter) noise images, and presented each image both at full size and windowed to a 1 degree diameter. Each line in this panel compares the MR for the preferred grating (left) and white noise images (right) for a neuron. White circle indicates the average MR across all neurons. White noise images result in much weaker suppression than preferred gratings (MR noise = 0.96; MR gratings = 0.26; p<0.0001), consistent with the predictions of our framework. (b) We also tested a second manner for removing higher-order statistical dependencies: we phase-scrambled natural images, which removes higher-order structure while maintaining the amplitude spectrum. For this manipulation, we chose a small subset (N=10) of the experimental images that were most often inferred homogeneous. We left the image inside the RF unaltered, but replaced the part of the image in the surround by its phase-scrambled version (Guo et al. Eur J Neurosci 21, 536-548, 2005). The phase-scrambling was achieved by taking the 2D Fourier transform of the image in polar coordinates, randomly shuffling the phase values, and taking the inverse Fourier transform back to pixel space. Each line in this panel compares the MR for a subset of natural images (N=10; left) with that of the same images after phase scrambling; white circles indicate the average MR across all neurons. Suppression was substantially weaker in the phase-scrambled images (MR phase-scrambled = 0.82; MR natural = 0.34; p<0.0001), consistent with the prediction of our framework.

Supplementary Figure 6 Distribution of phase-sensitivity index.

F1/F0 index was defined as the ratio of the amplitude of the first harmonic to the mean response across four phases (0, 90, 180 and 270), measured using the static grating whose orientation and spatial frequency evoked the strongest response. Values close to 0 indicate weak or no sensitivity to gratings phase (i.e. complex cells), whereas values of 1 or larger indicate phase-sensitive (simple) cells. Most cells showed little phase modulation, consistent with our use of a quadrature pair to represent the RF.

Supplementary Figure 7 Inference about image homogeneity based on measured neuronal tuning.

(a,b) Same as in Figure 6a,b of the main text, but representing each cell with a filter chosen based on tuning measurements rather than chosen to maximize fit quality. Specifically, we measured each cell’s tuning with static grating stimuli (described in Methods). The orientation preference of the neuron was defined by fitting a Von Mises function to the responses to small gratings presented at 16 orientations. The preferred spatial frequency and size of the RF were defined as the location of the peak of the respective measured tuning curves, for the grating orientation that matched the preferred orientation most closely. The filters were then chosen to match the orientation, spatial frequency and size preference. We included in our analyses all neurons for which the filter outputs agreed well with the orientation and spatial frequency tuning (R2 ≥ 0.5; n=41 neurons). (c,d) same as in Figure 7d,e of the main text, but for filters chosen based on measured neuronal tuning. Together, these analyses show that the importance of image homogeneity for explaining suppression does not depend in any way on fitting filters to responses—it is evident even when the neuron’s filters are measured directly with gratings. However, the stronger suppression for homogeneous images was contingent on correct estimation of the filters representing each neuron: the difference between image classes either vanished or changed sign when we used filters with a randomly chosen tuning profile, rather than those based on gratings measurements or fitting (not shown).

Supplementary information

Supplementary Text and Figures

Supplementary Figures 1–7, Supplementary Table 1 and Supplementary Modelling (PDF 1161 kb)

Supplementary Methods Checklist (PDF 448 kb)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Coen-Cagli, R., Kohn, A. & Schwartz, O. Flexible gating of contextual influences in natural vision. Nat Neurosci 18, 1648–1655 (2015). https://doi.org/10.1038/nn.4128

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/nn.4128

This article is cited by

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing