1 What is visible across the visual field?

3 Andrew M. Haun ([email protected]),

4 Center for Sleep and Consciousness,

5 Department of Psychiatry,

6 University of Wisconsin-Madison

7 Wisconsin, USA
 8 Abstract

9 In this paper I use a detailed model of human spatial vision to estimate the visibility of some perceptual

10 properties across the visual field, including aspects of colorfulness, sharpness, and blurriness. The

11 model is constructed to reproduce several patterns of human contrast sensitivity, functions of

12 contrast, scale and retinal eccentricity. I apply the model to colorful, complex natural scenes, and

13 estimate the degree to which color and edge information are present in the model’s representation

14 of the scenes. I find that, aside from the intrinsic drift in the spatial scale of the representation, there

15 are not large qualitative differences between foveal and peripheral representations of ‘colorfulness’

16 and ‘sharpness’.

17

18 1. Introduction

19 When looking at a scene one may get the feeling that their visual experience is colorful and sharp

20 across the full extent of the visual field. Even a savvy observer who knows about the higher objective

21 resolution of foveal vision, or about the regular, rapid, and involuntary movement of the foveae from

22 one part of the scene to another, is likely to get this feeling. However, it is frequently argued that this

23 impression is illusory. Such arguments – which tend to be part of larger philosophical or theoretical

24 accounts of vision (Cohen et al., 2016; Dennett, 2005; Lau & Brown, 2019; Schwitzgebel, 2008) -

25 are always claimed to derive from physiological or psychophysical facts about visual perception: cone

26 density declines with retinal eccentricity (Curcio et al., 1990), acuity declines even more rapidly and

27 more severely for color than for achromatic targets (S. J. Anderson et al., 1991), contrast sensitivity

28 is poorer peripherally than foveally for most spatial targets (Pointer & Hess, 1989; Robson & Graham,

29 1981), and declines more severely for color targets (S. J. Anderson et al., 1991). Therefore, it is

30 argued, however it seems (or seems to seem) to us, peripheral vision just cannot support the kinds
31 of color or sharpness percepts supported by foveal vision, and the only explanation for our subjective

32 feelings is that they are illusive.

33 I demonstrate in this paper that it is the argument-to-illusion that is mistaken: human spatial vision

34 permits near-invariance of color and sharpness representations across the visual field. This

35 invariance in representation is despite the real and significant changes in sensitivity across the visual

36 field. Crucial to this demonstration is that we take the view that color or sharpness are entirely

37 perceptual qualities, intrinsic to the observer - stimuli themselves have no such properties (stimuli

38 have other kinds of properties that can evoke the relevant perceptual qualities). Evidence for this view

39 of color includes the dissociation of stimulus properties from perceptual color experiences, such as

40 in color constancy (Gegenfurtner et al., 2015), lightness and brightness illusions (Adelson, 2000),

41 color afterimages and hallucinations (McCollough, 1965), color phosphenes (Dobelle & Mladejovsky,

42 1974; Grusser, 1995), and color in dreams (Kahn et al., 1962). Evidence for a corresponding view of

43 sharpness includes similar dissociations, such as in blur adaptation phenomena (Webster et al.,

44 2002), hallucinatory percepts with clear detail (Richards, 1971), sharp-textured phosphenes (Nebel,

45 1957; C. W. Tyler, 1978), and experience of clear percepts in dreams (Rechtschaffen & Buchignani,

46 1992).

47 To demonstrate the ‘intrinsic invariance’ of color and sharpness representations in visual

48 experience, I construct a detailed model of human spatial vision. The model is constructed to

49 reproduce known psychophysical patterns of human contrast perception. It embodies the same facts

50 that have sometimes been employed to make claims about degraded peripheral visual experience.

51 The results are therefore drawn as directly as they can be from empirical observations.

52

53 2. Results

54 2.1 Application of the model to natural scenes

 Figure 1. Some of the images used as stimuli for the contrast perception model.

55 The central question in this study can be put this way: given what we know about human contrast

56 sensitivity, what can a typical human observer see in a colorful natural scene? To answer this question

57 I used 100 colorful natural scene photos (Figure 1) as stimuli for the model (see Section 4). The

58 scenes were all pulled from the ‘Panorama’ section of the Wikipedia directory of featured pictures

59 ((Various), 2019); these images are generally very high resolution, minimally-compressed, and full-

60 color, and they are of the kinds of interesting scenes that might elicit naïve claims about the apparent

61 vividness of a visual experience. The main selection criterion was that each image must have height

62 and width equal to or greater than 1536 pixels. Inputs to the model were cropped to 1536x1536

63 pixels; if an image had least dimension greater than 3072 pixels, it was cropped down to the nearest

64 multiple of 1536 and then resized down to 1536x1536. Some subjective criteria were applied in

65 selecting the scenes, including that the central region of the scene should contain some more

66 ‘interesting’ content than just ground (or sea) and/or sky; some content should be ‘near’, i.e. obviously
67 telescopic imagery was excluded; images should seem colorful (scenes like pictures of the desert or

68 of snowy mountains that seemed effectively monochromatic were generally not included, though

69 some were); and a rough balance was sought between ‘natural’ and ‘artificial’ scenes (i.e. of scenes

70 with and without obvious human influence). The list of source images (urls and photographer credits)

71 is provided in the supplementary materials. I did not make any effort to match the ‘true’ visual angle

72 of the scenes to the visual angle of the model’s visual field – the necessarily information to recover

73 the true angle was not generally included with the images.

74 To ‘view’ the selected scenes, the model was given a 32° square visual field (which would fit

75 comfortably within the angle of view of photographs not taken with wide-angle or long-focus lenses),

76 extending from 1° left of the fixation point to 31° right; and from 16° below to 16° above. The resolution

77 of the field was 1536x1536 pixels, so the vertical/horizontal Nyquist limit for the pre-filtering stage of

78 the model was 24 cycles per degree. Under these conditions, the smallest filter in the model (for the

79 foveal luminance-contrast channel) had peak sensitivity to 10.7 cycles per degree, so the model was

80 relatively (though not perfectly) insulated from sampling artifacts. The model’s response (in the form

81 of the maximum cross-frequency response at each field location) to one stimulus image is shown in

82 Figure 2. Right away we learn something about natural scenes: they are composed of high contrasts,

83 as far as the visual system is concerned. Detection thresholds are routinely exceeded across the

84 model visual field, as shown by the very-high d´ values elicited across the scene. If we were doing

85 signal detection experiments with the image components evoking these responses, a ‘no-lapse’

86 observer would respond perfectly across thousands of trials. Figure 2 also shows the cross-frequency

87 maximum for each color channel; suprathreshold responses are consistent across the field, with the

88 blue-yellow channel clearly the least sensitive.

 Figure 2. Top: stimulus image exposed to the model. Middle: model responses to the stimulus, collapsed
across scale, orientation, and color (using the M-norm for scale and winner-take-all for orientation and
color). Bottom: The ‘visible image’ according to the model response. This image is composed only of
image contrasts that elicit ‘suprathreshold’ responses.

89 Our original questions concern whether the content of natural scene experiences is rightly

90 characterized as ‘colorful’ and ‘sharp’ across the extent of the visual field. So, how do we evaluate

91 these two visual qualities? Colorfulness is the more straightforward to address, so we start there.

92

93 2.2 Colorfulness

94 ‘Colorfulness’ is an informal term, but in its informal usage it is intended to imply either variegation

95 (i.e. many distinct hues) or vividness (i.e. the presence of a highly saturated color) of colors, or

96 especially both. Here I used the ‘hue-saturation-value’ (HSV) representation of color content to

97 capture these features (Smith, 1978).

 98

99 2.2.1 The “visible image”

100 HSV is a pixelwise representation of image content, whereas the spatial vision model generates

101 a higher-dimensional matrix of signal-noise ratios. However, it is straightforward to translate the

102 model’s response to an image back into the form of a “visible image”, where we can make use of

103 HSV. We transform each filter’s signal-noise ratio (d´) into the range [0,1], using this value to weight

104 the positive cosine phase of the filter, and adding all the weighted filters to an output image. The

105 appropriate transformation of d´ is the ‘accuracy’ or ‘reliability’: the greatest difference between the

106 yes/no hit-rate and false-alarm rate:

𝑅 𝑅
107 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝐻𝑅 − 𝐹𝐴 = Φ ( 2 + C) − Φ (− 2 + 𝐶) (1)

108 Here Φ is the cumulative normal distribution function, R is the d’ elicited by the stimulus, and C is the

109 decision criterion. This procedure maps near-zero d´ to near-zero accuracy, and higher d´ to

110 accuracy approaching one. The output image is then composed of content only to the degree that it

111 elicits psychophysically accurate responses. That is, if a filter response is highly likely to have been

112 elicited by a stimulus (a hit), and highly unlikely to have been due to intrinsic noise (a false alarm),

113 then its contrast is considered ‘visible’ and composes a part of the output image. This statistic is

114 sensitive to the criterion C that divides ‘yes’ and ‘no’ responses: I assume an optimal (accuracy-

115 maximizing) criterion of C=0 in these analyses. It can be shown that variation in criterion does not

116 change the pattern of the results (varying C in either direction only reduces Accuracy).

117 The reconstructed image is in CIElab coordinates; we transform this to HSV coordinates (chaining

118 the MATLAB lab2rgb and rgb2hsv functions) and proceed with the colorfulness analysis.

119

120

121
122 2.2.2 Sampling across the visual field

123 An example of a ‘visible image’ is shown at the bottom of Figure 2. The image seems similar to

124 the original, except for its reduced peripheral resolution. Its colorfulness doesn’t seem reduced

125 across the model field´s extent, but the eye is the wrong judge here (the ‘double-pass problem’: cf.

126 (Peli, 1996)). This is where we make use of the HSV representation, and sample hue variation and

127 saturation across the field. What is the right way to do this? A human observer making a judgment

128 about some image property across the visual field must be using spatial attention, and it is known

129 that the size of the spatial attention ‘spotlight’ varies with eccentricity, with a radius r following a

130 pattern much like the scaling of acuity (Intriligator & Cavanagh, 2001):

131 𝑟 = 𝑘0 (1 + 𝐸⁄𝐸 ) + 𝑘𝑐 (2)

2

132 Here I followed Intriligator and Cavanaugh’s results (Intriligator & Cavanagh, 2001) and set k0 to 5

133 minutes of arc, and E2 to 0.34 degrees eccentricity; but since this results in model foveal windows

134 just a few pixels across (which would result in severe undersampling at the fovea of local quality

135 values for the measures described below) a constant kc of 1 degree was added at all eccentricities

136 (equivalent to increasing both the original constants by one factor). This eccentricity-scaled sampling

137 rule should still reflect the constraints on a human observer’s spatial sampling strategy, if they are

138 tasked with investigating the local spatial distributions of some visual quality like color.

139 One could argue that the scaled sampling rule is biased with respect to the central research

140 question, so the scaled rule was compared with an unscaled sampling rule, fixing r at 3.75 degrees.

141 This is the median (from 0 to 31 degrees eccentricity) of the scaled sample regions, and is about the

142 size of the parafoveal region. (It is also the radius of the round window made by touching the tips of

143 my thumb and forefinger at a viewing distance of about 40 cm.)

144
 Figure 3. Colorfulness statistics as a function of eccentricity, for eccentricity-scaled (A) and fixed (B)
sample areas. C,D) Saturation distributions. Each gray dot is the mean HSV saturation for one image,
within the sample area. The black lines trace the average upper quantiles (0.5, 0.75, 0.9, 0.95, and
0.99) over all images. E,F) Hue entropy. Each gray dot is the average hue entropy for one image,
within the sample area. The black lines trace the average entropy over all images. The vertical bars
are upper and lower entropy quartiles (0.25, 0.75) over all images.

145 2.2.3 Colorfulness over eccentricity

146 With these sampling rules, we assess colorfulness of model-visible images at each position (in 1-

147 degree steps) along the horizontal midline as shown in Figure 3. Two measures capture colorfulness

148 at any sample position: first, there is the distribution of saturations, with high saturation quantiles

149 reflecting the most colorful parts of the sample; second, there is the hue entropy, which reflects the

150 variegation of a sample (how many different hues are encountered there). Intuitively, the hue entropy

151 should be computed for saturations that produce visible colors – for this demonstration a relatively
152 low bar of 0.2 saturation was set, since a higher bar tended to reduce sample sizes to zero for many

153 scenes, even near the fovea (high-saturation colors in natural images are relatively rare). Saturation

154 quantiles are self-explanatory, and hue entropy (H) is defined specifically as:

155 𝐻 = − ∑ℎ𝑢𝑒𝑠 𝑝(ℎ𝑢𝑒|𝑠𝑎𝑡 > 0.2) log 2 𝑝(ℎ𝑢𝑒|𝑠𝑎𝑡 > 0.2) (3)

156 The distribution p(hue | sat>lim) was defined over 256 HSV hue bins. Maximum entropy – an even

157 distribution of hues across the full range – would be H=8 bits. Concentration of color around particular

158 hues appears as lower entropy.

159 Figure 3CD shows the upper quantiles of the saturation distribution as a function of eccentricity

160 for the different sampling rules; Figure 3EF shows the hue entropy. For the scaled rule, there is little

161 dependence on eccentricity of the saturation distribution. For the fixed rule, there is a gradual decline

162 with eccentricity in the saturation of the highest quantiles. The relationship between eccentricity and

163 hue entropy is similarly muted: entropy increases slightly with eccentricity for the scaled rule (for the

164 scaled rule the average increase in entropy is about .01 bits per degree), and decreases slightly for

165 the fixed rule (on average losing about .04 bits per degree). Entropy is generally around 4 bits in the

166 investigated range.

167 Considering that the fixed size sampling rule is unrealistic and probably perceptually impossible,

168 the slight declines in colorfulness for that rule are not what we should expect a human observer to

169 report. At the same time, the attention-based rule might impose a too-rigid frame to visual field

170 sampling (despite its relative ‘flattening’ by adding 1 degree of radius at each eccentricity): perhaps

171 observers (especially expert observers) are, with some effort, able to attend much smaller zones in

172 the periphery and much larger zones nearer the fovea, when they are trying to “sample fairly” across

173 the visual field. So the scaled rule might likewise overestimate the relationship in the opposite

174 direction. Left with a relationship somewhere in between a slight increase and a slight decrease, it is
175 reasonable to describe the relationship between visible colorfulness and retinal eccentricity as very

176 small or negligible.

177

178 2.2.4 Color in the response image

179 The preceding analysis relies on the visible image transformation of the model responses. A

180 similar analysis could be performed on the ‘raw’ response image. For example, a saturated red spot

181 in the visible image is, for the model, a vector of channel responses at that visual field location: across

182 scale and orientation, red-green channel filters with ‘red-on’ phase will have suprathreshold

183 responses, as will achromatic ‘dark-on’ phase filters; meanwhile, filters in the blue-yellow channel will

184 have weakly yellow-on responses. If there is a ‘saturated red’ HSV triplet at different eccentricities,

185 this implies that a similar response vector exists at each eccentricity. Each unique HSV triplet

186 corresponds to a unique set of such vectors (allowing for variation in scale and/or orientation of the

187 responding filters, according to the shape of the colored spot). The visible image transform captures

188 the same structure, as well as the decisional aspects of the responses, and so for simplicity the

189 underlying response vectors are not presented in this paper.

190

191 2.3 Sharpness

192 There is no standard pixelwise element of “sharpness” analogous to hue and saturation. What

193 could be the response image correlate of apparent sharpness? Let us define this property as the

194 absence of apparent blur: if a feature is seen but does not appear blurry, then it appears sharp. The

195 spatial spread of ‘just detectable blur’ increases in proportion to eccentricity in a similar way as acuity

196 (Hess et al., 1989; Levi & Klein, 1990; Maiello et al., 2017; B. Wang & Ciuffreda, 2005). That is, it is

197 approximately true that, across the visual field, if the spatial spread of blur is less than the acuity limit,

198 a ‘blur percept’ will not be evoked; but if the spread is larger, it will be.
 Higher-resolution edge Lower-resolution edge

Finer filter

scale fineness

scale fineness
scale
(‘foveal’) ✓ ✗
✓ ✗

filters
✓ ✓
✓ “sharp” ✓ “blur”
(complete (incomplete
response) response)

edge
spatial position spatial position

Coarser filter
scale fineness

scale fineness
scale
(‘peripheral’)

✓ ✓
✓ ✓
✓ “sharp” ✓ “sharp”
✓ ✓

spatial position spatial position

Figure 4. The ‘scale space’ conception of edge sharpness and blur. The left panels illustrate a high-
resolution edge (the vertical black line), which exists across a wide range of scales, extending to very
fine scales. The right panels illustrate a low-resolution edge, which disappears at finer scales. A fine-
scale ‘foveal’ filter system, illustrated by the circles in the upper panels, can distinguish between these
two features: for the lower-resolution edge, the finer two filter scales do not respond. The coarse-
scale ‘peripheral’ filter system does not distinguish between the two features. However, notice that
there is nothing intrinsically that distinguishes the two phenomena on the left, and lower-right: all three
are cases where there is a complete response. Each of these registers as a ‘sharp edge’.

199

200 2.3.1 Scale space, sharpness, and blur

201 A measure applicable to the multiscale channel contrast responses of our model can be derived

202 from the “scale space” notion of feature representation (Georgeson et al., 2007; Koenderink, 1984;

203 Perona & Malik, 1990; Witkin, 1987). Figure 4 uses this concept to illustrate the distinction between

204 image resolution (i.e. “physical sharpness”) and apparent sharpness (“perceptual clarity”), and to

205 explain how we can find a correlate of apparent sharpness in the spatial vision model. These
206 properties (physical and perceptual sharpness) may often be conflated, but they are just as distinct

207 as reflection spectrum and perceived color. The left two panels show the scale-space representation

208 of a high-resolution edge: such an edge exists, in physical terms, as a feature at a particular spatial

209 location (on the x-axis) that extends from coarse scales up to rather fine scales. The right panels

210 show a low-resolution edge: this edge exists as a feature that extends from coarse scales up to only

211 moderately fine scales. These are two physically different features, but they do not determine

212 perceptual qualities: perceptual mechanisms also have to be taken into account. The upper and lower

213 panels contrast two different filter scales – a finer ‘foveal’ filter set, and a coarser ‘peripheral’ set. The

214 foveal filters distinguish between the two features in that the lower-resolution edge doesn’t elicit any

215 response from the smallest filters. So, the high-resolution edge elicits a ‘complete’ filter response,

216 and the low-resolution edge elicits an ‘incomplete’ response. The peripheral filters do not distinguish

217 the two features: for this filter set, both edges elicit ‘complete’ responses.

218

219 2.3.2 Filter completeness as sharpness

220 The implication of the scale-space demonstration is that ‘apparent sharpness’ is closely related

221 to a complete filter response, and is therefore distinct from the physical resolution of a stimulus. This

222 link between apparent sharpness (and blur) and a cross-scale response has been proposed many

223 times, though always in different forms (e.g. (Elder & Zucker, 1996; Georgeson et al., 2007; Z. Wang

224 & Simoncelli, 2003)), and usually either explicitly or implicitly with reference to foveal perception (one

225 exception is (Anstis, 1998)). Going by this conception of apparent sharpness, we can recruit the

226 ‘response accuracy’ statistic of Eq.1 and define apparent sharpness as filter completeness, i.e. the

227 cross-scale product of accuracies at a given location:

228 𝐹𝐶𝜃 = ∏𝑓 Accuracy(𝑅𝑓,𝜃 ) (4)

 Figure 5. A) The visible image of a scene’s luminance contrast. B) The regions of the scene where
responses are filter complete (FCθ >0.96) are colored cyan. C) The proportion of FCθ >0.96 pixels
(‘Edge density’) within the sampling window positioned at each horizontal eccentricity. Gray lines are
measures for each scene (N=100), black lines are the mean and upper/lower quartiles over all scenes
for the scaled sampling rule. Dotted lines are for the fixed sampling rule (they do not differ appreciably
from the scaled rule). The average was not taken for windows extending outside the model field.

229 We could require further that the cross-frequency response obeys certain phase correlations (as for

230 example in the perception-inspired models of (Portilla & Simoncelli, 2000; Z. Wang & Simoncelli,

231 2003)), but for generality we require simply that a feature at some location is perceptually sharp when

232 there is a highly-accurate response at all filter scales at that location, i.e. when for some orientation θ

233 the quantity FCθ is near one.

234 Figure 5 illustrates the application of this idea to the model responses to natural scenes, taking

235 filter completeness to be FCθ values greater than 0.96 (allowing that each of four filter responses has

236 accuracy ~0.99)1. Here we evaluate filter completeness only for the luminance channel, since its

237 smaller filter size means that it must in any case be the driver of sharpness judgments. The first panel

238 (5A) shows the visible luminance-contrast image for a particular scene; the next panel (5B) highlights

239 the regions of the image where the model response was ‘filter complete’. Note that the image contrast

240 was attenuated towards the edges of the scene, so that image boundaries would not contribute to

241 the edge analysis. Using the same two sampling rules as in the colorfulness analysis, the last panel

242 (5C) shows the average filter completeness – the mean of FCθ >0.96 within the sample region - as a

243 function of (horizontal) eccentricity. This analysis is insensitive to sampling rule, but there is a clear

244 positive trend with eccentricity of increasing filter completeness. Under the plausible hypothesis that

245 filter completeness is apparent sharpness, the spatial vision model does not support the notion that

246 apparent sharpness should decline with eccentricity.

247

248 2.4 Apparent blur of visible content

249 Apparent sharpness does not capture the whole story: if a feature does not appear sharp, then it

250 (by definition) must appear blurry, and there is plenty of room for features to appear more or less

251 blurry depending on various circumstances. Apparent blur, i.e. how blurry something appears to be

252 (with ‘sharpness’ being the minimum of apparent blur), is usually measured by perceptually matching

1
Why not tie apparent sharpness simply to the responsiveness of the smallest filter at each position? There
are some image features, mainly fine-grained textures, that will not be judged as ‘sharp’ under the FC metric,
but would be if the metric were relaxed to extend only over the finer (or finest) filter scales. However, the
model would then be inclined to judge features as sharp erroneously, such as when white noise is added (as
in (Kayargadde & Martens, 1996; Kurihara et al., 2009)), even though adding white noise to edges in natural
scenes tends to reduce their apparent sharpness ((Kurihara et al., 2009); perhaps as a consequence of
‘coarse-scale suppression’ as reported in (Haun & Hansen, 2019; Haun & Peli, 2013b)). Considering these
issues, the broadband FC metric is the more conservative choice: it should underestimate, rather than
overestimate, apparent sharpness.
 Figure 6. A) Stimuli used for the replication of Galvin et al (1997).11 levels of gaussian blur were
applied to a monochrome vertical edge. The stimuli are illustrated to scale: the image is 10.67 degrees
wide. B) The ‘apparent blur’ statistic β for a simple edge stimulus convolved with a gaussian with the
scale constant on the x-axis (scale constant is in minutes of arc, i.e. 1°/60). The apparent blur metric
is explained in the text. Each line is a Gaussian fit to the blur metric as a function of stimulus blur, for
each of six stimulus eccentricities. C-G) Apparent blur matched between a fixed-blur peripheral
stimulus and an adjustable-blur foveal stimulus. Model matches between the foveal and peripheral
blurs were computed numerically using the gaussian curves fitted in B. Veridical matches would be
on the main diagonal, and matches above the main diagonal mean that the peripheral stimulus
appears sharper (less blurry) than it would if it were viewed foveally. Symbols are data from Galvin et
al (1997)’s first experiment (replotted from their Figure 2). As eccentricity increases, the model
becomes less accurate, over-estimating perceived sharpness of peripheral content. The model
judges peripheral edges to be even sharper than the human observers judged them to be. Over the
five test eccentricities {8.3, 16.7, 24, 32, 40} the average difference between model and data is small
but consistently negative: {0.2, -1.0, -1.1, -2.0, -1.9} arcmin, respectively.

253 the blur of one percept to the blur of another. In a pair of studies especially relevant to the central

254 question of this study, Galvin et al measured apparent blur matches between foveal and peripheral

255 stimulus positions (Galvin et al., 1997, 1999). They found that a blurred edge viewed peripherally was

256 subjectively matched to a less-blurred edge viewed foveally (scattered symbols in Figure 6). They
257 called this effect “sharpness over-constancy”. They proposed that some mechanism must exist that

258 corrects (or over-corrects) for the lower resolution of peripheral vision. In their view, the peripheral

259 stimuli appeared sharper than they should (implicitly they were taking ‘foveal appearance’ as the

260 standard for how things ought-to-look). The spatial vision model suggests a different interpretation of

261 their results. Figure 6 replots Galvin et al’s data (from their Experiment 1), along with perceived-blur

262 matching functions from the spatial vision model. ‘Apparent blur’ does not have an obvious

263 implementation so I adapted the simple ‘response slope’ model of (Elliott et al., 2011; Haun & Peli,

264 2013a), where apparent blur is equated to the rate of decrease (m) of the perceptual response (here

265 R) as the log filter scale decreases (or as log frequency f increases):

266 R(𝑓) = 𝑚 ln 𝑓 + 𝑏 (5)

267 This model is designed to explain perceived blur of a narrow class of stimuli (blurred by steepening

268 the amplitude spectrum slope; (Elliott et al., 2011; Haun & Peli, 2013a)), and it performs very badly

269 (i.e. non-monotonically for increasing blur levels) for gaussian-blurred stimuli. However, I found that

270 adjusting the slope term by the local response (the M=4 norm 𝑅̅ ; see section 4.7) yields blur estimates

271 that monotonically increase with stimulus blur (Figure 6a), so the ‘apparent blur’ term is:
𝑚
272 β= (6)
𝑅̅

273 When there is similar response across filter scale (as would be ideally evoked by the large edge stimuli

274 of the Galvin et al study), β will be near zero – when response declines with increasing filter scale, β

275 will be negative. β can also run to positive values when there is only fine-scale content like a fine-

276 grained texture.

277 This perceived-blur model (as with most or all others proposed) has been tested only with foveal

278 psychophysics data, and it fits the matching Galvin data only roughly (Figure 6C-G). However, notice

279 how it fails: held to a foveal standard, the spatial vision model behaves as though peripheral content,

280 especially at larger eccentricities, is perceived as even less blurry than Galvin et al found it to be
 Figure 7. A) The ‘apparent blur’ measure over
eccentricity for the scene stimulus set. Unlike
the simple edge stimuli of Figure 6, natural
scenes typically evoke positive apparent blur
scores, which we may interpret as ‘apparent
sharpness’. In general, regions of a natural
scene at optical infinity (as in our stimulus set)
will evoke these positive scores; negative
scores generally correspond to featureless
regions, usually sky, where most content is in
a very low frequency gradient.

281 (Figure 6E-G). That is, rather than sharpness over-constancy, there may be an under-constancy at

282 work in human peripheral vision. One possible culprit here (aside from deficiencies of the model) is

283 crowding (Rosenholtz, 2016): a subject’s relative inability to judge the qualities of content in peripheral

284 vision, despite the psychophysical availability of the necessary information, might contribute to

285 judgments of apparent blur. It is also possible that Galvin et al’s experiment confounded, to some

286 degree, foveal and peripheral judgments of sharpness and blur, by not perfectly controlling for eye

287 movements – a peripheral blur stimulus, if moved even briefly closer to the fovea, would elicit a

288 “blurrier” judgment than if it had not been moved. If a few fixational errors occurred over the course

289 of the experiment, the result would be a compression of the matching functions towards the identity

290 line. Finally, it could be that observers have some natural knowledge that the objective resolution of

291 peripheral vision is less than that of foveal vision, and they are injecting that knowledge into their

292 decisions about apparent peripheral blur. At any rate, the spatial vision model does not predict that

293 peripheral stimuli should be judged as blurrier than foveal stimuli, or indeed as blurrier than human

294 observers themselves tend to judge them.

295
296 2.4.1 Apparent blur and sharpness of natural scenes

297 The apparent blur model of the previous section is straightforward to apply to the natural scene

298 contrast responses underlying the analyses in previous sections. This analysis requires the obvious

299 caveat that, as shown in Figure 6, the apparent blur model is a rough fit to the one available dataset

300 (Galvin et al., 1997). Also, except for the example of Figure 6, the model has never been validated

301 on local image patches, only on ‘whole image’ statistics (Elliott et al., 2011; Haun & Peli, 2013a).

302 However, the model is qualitatively accurate (the model is not that far off as shown in Figure 6 –

303 it closely matches data at smaller eccentricities, and misses by at most 2 minutes of arc at larger

304 eccentricities, although this is proportionately a large deviation from the data - and is at least

305 monotonic with the psychophysical patterns). Also, as suggested in the previous paragraph, it is

306 possible that the model is more accurate than it seems, and that methodological flaws or innate

307 biases resulted in the human observers slightly overestimating peripheral blurriness.

308 With these caveats in mind, consider how the apparent blur parameter β, as evoked by the scene

309 images, varies with eccentricity: that is, it does not vary much at all (Figure 7), averaging a positive

310 value at every eccentricity. If, in viewing a panorama, normal human observers are comparing some

311 statistic like β across their visual fields, they should find that the distribution of apparent blurs is not

312 obviously dependent on retinal position. In fact, if the intrinsic blur statistic is anything like β, they

313 should find that a typical scene (albeit one viewed at optical infinity, as in most of our sample scenes)

314 does not contain much blur at all.

315

316 2.5 Attention

317 The target data on which the model is constructed and tested were all collected under conditions

318 where the target stimulus was the sole focus of the observer’s spatial attention. So, my findings would

319 seem to hold for, at a minimum, judgments about attended color and sharpness qualities. What
320 happens to colorfulness and sharpness when spatial attention is withdrawn from a region of the visual

321 field?

322 The effects of spatial attention are complex (involving shifting tuning for individual neurons,

323 changes to perceptual organization and recognition, and surprising phenomena like inattentional

324 blindness) and, in general, its mechanisms are poorly understood. However at the level of early spatial

325 vision, there is some clarity about what is happening. Neural and psychophysical measures seem to

326 agree that spatial attention corresponds to enhancement of contrast response (Bressler et al., 2013;

327 Buracas & Boynton, 2007; Guo et al., 2020; Herrmann et al., 2009; Reynolds & Heeger, 2009); so,

328 withdrawal of spatial attention means reduced contrast sensitivity, reduced perceived contrast (but

329 see (Schneider & Komlos, 2008)), and attenuated neural response. A crude implementation of this

330 enhancement (or of its withdrawal) in the model would be to vary the amplitude (Rmax in Eq.9) of the

331 contrast response function. If the main model reflects the enhanced response state of an attended

332 region, we can implement the withdrawal of attention by reducing Rmax. Reducing this parameter

333 would mirror the kinds of reductions seen in neural contrast response functions (Buracas & Boynton,

334 2007; Gandhi et al., 1999; Luck et al., 1997), and would also reduce contrast sensitivity and

335 perceived contrast judgments (assuming that perceived contrast is strongly linked to contrast

336 response magnitude) in similar ways to what is observed psychophysically (Carrasco et al., 2000,

337 2004; Carrasco & Barbot, 2019; John M. Foley & Schwarz, 1998; Huang & Dobkins, 2005).
 Figure 8. Simulating the effects of attention on visibility of color and sharpness. A) Attended (Rmax=30)
and unattended (Rmax=22.5) contrast response functions for a range of z-values (threshold
parameters). B) Unattended (solid line) contrast sensitivity is reduced relative to attended (dashed
line) sensitivity. The sensitivity curve here is for foveal luminance contrast before the linear gain
parameter (gach). C) Plotted as threshold elevation, the difference between unattended and attended
thresholds tends to around 20%. D) There is some reduction of saturation for the unattended
condition. E) Hue entropy is not affected by the attention manipulation. F) Edge density is reduced in
the unattended condition. G) Apparent sharpness (positive values of the apparent blur metric) is
increased by inattention. The dotted line shows an adjusted metric that assumes knowledge of the
inattentional reduction in response amplitude (R max).

338 I re-ran the main analyses of colorfulness and sharpness with a version of the model with Rmax

339 reduced by 25% (Figure 8A, Rmax=22.5; this is consistent with the attentional effect on fMRI BOLD

340 response (Buracas & Boynton, 2007; Gandhi et al., 1999; Guo et al., 2020); the relation of BOLD

341 contrast response to perceptual contrast is not a straightforward equivalence, however). This is a

342 significant reduction that produces psychophysical effects in a similar range to what has been

343 observed in numerous studies, reducing sensitivity by around 20% (Fig.8B,C), depending on the

344 default sensitivity of each mechanism, but I did not try to fit the reduction to any particular dataset

345 (see (Carrasco et al., 2000; Lee et al., 1997) for similar effects, and many other effects of similar

346 magnitude reviewed in (Carrasco, 2011)).

347 Figure 8 also compares the original perceptual quality measures of the ‘attended’ scenes with

348 measures of the ‘unattended’ scenes. The changes to colorfulness are modest: unattended regions

349 have reduced saturation (99th percentile saturations, on average, drop from 0.54 to 0.51; Fig.8D) and

350 marginally-reduced hue entropy (from 4.31 to 4.25 bits; Fig.8E). The changes to sharpness are more

351 significant (Figure 8F): edge density (averaged over eccentricity) drops from around 0.27 to around

352 0.21, roughly proportional to the change in response amplitude. Interestingly, the apparent blur

353 metric (β) increases slightly with inattention (Fig.8G) – while reducing Rmax would reduce the slope

354 estimates underlying the blur metric, the normalizing factor, being reduced by the same factor,

355 compensates for the reduction. If we think that apparent blur should change similarly to edge density,

356 we can suppose that the visual system ‘knows’ about the inattentional reduction of R max, and takes

357 this reduction into account by reducing R-based statistics by the same proportion: basically, we

358 multiply the original β by the reduction factor. The dotted line in Figure 8G shows this adjusted β is

359 similarly reduced to the reduction in edge density.

360 These attentional effects on visible qualities are not very dramatic, but they are real. If the actual

361 effects of attention on contrast sensitivity are larger than what is modeled here, then the effects on

362 perceptual qualities would be correspondingly larger. Overall, this may support a weak version of the

363 so-called ‘refrigerator light illusion’ (Block, 2001), i.e. the notion that unattended properties of visual

364 experience may be somewhat different from attended properties – but that we wouldn’t notice the

365 difference, since whenever we check (using attention) we find the attended properties.

366 It is worth considering why, if the contrast response (and sensitivity) is changed so significantly,

367 are visible qualities not more dramatically affected? Saturated colors or sharp details are evoked by

368 high physical contrasts that yield (in the attended case) very high signal-noise ratios; halving these

369 ratios will generally still result in a suprathreshold response (e.g. going from d’=8 to d’=4). If halving

370 the contrast response results in a large decline in the accuracy of a feature, then the attended
371 response must already have been rather low (e.g. from d’=2 to d’=1). In terms of accuracy (Eq.1),

372 the reduction of contrast response magnitude only has significant consequences for features whose

373 attended response was in the range [0,~4]. Responses in this range already contribute only

374 marginally to colorfulness and sharpness (this is almost by-definition: of course low contrasts do not

375 contribute much to judgments of sharp edges or vivid color). The effects of attentional enhancement

376 of contrast response are on the faint and hard-to-see, rather than on the vivid and easy-to-see.

377

Figure 9. A summary of the main results. Color qualities and sharpness qualities, as assessed
against contrast responses to colorful panoramic scenes, do not differ dramatically between
parafoveal and peripheral visual fields. The values here are averages of the plots shown in Figures
3, 5, and 7, the arithmetic means of values below 5 degrees (for a ‘parafoveal’ measure) and
above 15 degrees (for a ‘peripheral’ measure).

378 3. Discussion

379 According to standard techniques for measuring human vision, basic capacities of visual

380 perception (sensitivity and resolution) decline significantly with increasing retinal eccentricity. These

381 facts have led some to conclude that perceptual qualities must therefore degrade with eccentricity.

382 In this paper, I have demonstrated that this conclusion is wrong: given the sensitivity and resolution

383 of the normal human observer, one would expect perceptual qualities to be rather stable across the

384 visual field (Figure 9). This demonstration requires only that we take an intrinsic perspective on spatial
385 vision: that the visual system can only know about what it can represent; it can’t know about what it

386 can’t represent. This idea was expressed particularly well by Anstis in his 1998 paper on how to

387 understand peripheral acuity:

388

389 “Why does the whole visual field normally look equally sharp all over, when there is clearly an

390 enormous degradation of the visible detail in peripheral vision? This is an ill-posed question.

391 After all, if our acuity were 1000 times better than it is, we could throw away our microscopes,

392 yet our 'limited´ foveal acuity which prevents us from seeing bacteria with the naked eye never

393 looks like any kind of subjective blur! The same applies to our limited peripheral acuity. A

394 channel cannot signal anything about stimuli to which it is not tuned, so peripheral retinal

395 channels must remain silent in the presence—or absence—of high spatial frequencies to

396 which they are blind.” (Anstis, 1998)

397

398 Despite the straightforwardness of this way of understanding visual perception, an ‘extrinsic’

399 perspective on vision may be closer to the mainstream of cognitive science, since it fits well with

400 overarching theories of computation and information-processing, and notions of veridicality and

401 intentionality. Under the extrinsic computational perspective, it may seem intuitive that, because there

402 are small image features that can be seen foveally but not peripherally, this means that peripheral

403 vision is actually blurry in comparison to foveal vision. The fact that it does not seem this way then

404 produces an excitingly counterintuitive thesis: that the visual field doesn’t really feel the way it seems

405 to feel, and introspective judgments about visual qualities are not to be trusted.

406 It has also been pointed out many times in the past that color perception, when considered

407 separately from size perception, is relatively independent of retinal position (Block, 2007; Haun et al.,

408 2017; Noorlander et al., 1983; Christopher W. Tyler, 2015). One rejoinder to this is that objects in
409 natural scenes, as opposed to eccentricity-scaled experiment stimuli, do not change size with eye

410 movements, and so the size dependence of color perception across the retina is not relevant to

411 natural vision (see responses in Burnston, 2018; or Van Gulick, 2007). However, natural scenes are

412 scale invariant (Burton & Moorhead, 1987; Ruderman & Bialek, 1994), meaning that, on average,

413 any location within a scene can contain spatial content at all scales. So, a filter complement of any

414 scale can expect to find content. Other things being equal, a neuron with a large color-opponent

415 receptive field in the periphery is as likely as one with a small foveal receptive field to find a stimulus

416 that drives it.

417

418

419 3.1 Conclusions

420 The spatial vision model I have used in this paper is not unique. There are many, many alternate,

421 and probably more elegant, formalisms for sensitivity across the visual field (e.g. (Watson, 2018;

422 Watson & Ahumada, 2016)), sensitivity to different levels of contrast (e.g. (Lu & Dosher, 1999)), color

423 vision, and other aspects of pattern vision. I selected the components of this model for their

424 convenience and compatibility (I do not believe that a comparable ‘omnibus’ model of contrast

425 sensitivity has been composed before). The important thing for my purposes is that different models

426 are generally psychophysically equivalent. That is, I expect that an alternate spatial vision model could

427 be constructed, but as long as it describes the psychophysical patterns listed in points i-x (section 4),

428 the ensuing statistical analyses should be the same. I do not believe that the results of this study are

429 the consequence of some peculiar modeling choices.

430 I hope it is clear that the scope of these findings is limited. What I find is that the familiar

431 psychophysical patterns of exponential decline in sensitivity with eccentricity and spatial frequency,

432 and the steeper decline for chromatic channels, do not determine the incapacity of peripheral vision
433 for representing color or sharpness. This is not to say that capacity for visual qualities must be the

434 same across the visual field. Crowding is an obvious difference between foveal and peripheral vision,

435 but how colorfulness or sharpness judgments are affected by crowding has not been a target of many

436 (or any) studies. Attention is another obvious difference, since it naturally resides at the foveal field,

437 but this has already been addressed. And there are many other higher-level visual processes that

438 can alter spatial vision qualities depending on context (two highly interesting examples related to

439 appearance of color or details in complex scenes across the visual field: (Balas & Sinha, 2007; Otten

440 et al., 2017)). But to my knowledge, no such higher-level processes are believed to prize foveal over

441 peripheral vision when it comes to judgments about sharpness or colorfulness.

442 So, on the basis of known psychophysical facts about human spatial vision, I conclude that a

443 normally-sighted human observer who claims to have a clear and colorful experience across her

444 visual field is justified in her claim by existing empirical evidence. Introspection about low-level visual

445 qualities might be pretty good after all!

446

447 4. Methods

448 The contrast perception model was implemented in MATLAB (code is provided). Empirical data

449 used to constrain model parameters, or for other purposes (e.g. the perceived blur analysis in Section

450 2), was extracted from the original study publications using the online WebPlotDigitizer tool (Rohatgi,

451 2017).

452

453 4.1 Core patterns of spatial vision

454 Listed below are the psychophysical patterns that I take to be most pertinent to the question of

455 color/sharpness perception across the visual field (some ‘exemplar’ references are provided for each

456 – all of these patterns have been observed in numerous studies).

457 i. Contrast sensitivity is relatively independent between three opponent chromatic axes (the three

458 ‘color channels’: luminance-contrast, blue-yellow contrast, and red-green contrast). That is,

459 interactions between same-channel patterns are much stronger than interactions between

460 different-channel patterns (Buchsbaum & Gottschalk, 1983; Krauskopf et al., 1982; Kathy T.

461 Mullen & Sankeralli, 1999).

462 ii. Spatial frequency acuity (the spatial frequency fa at which contrast sensitivity is = 1, i.e. minimal)

463 for achromatic and achromatic targets is inversely proportional to eccentricity plus a constant

464 (E2):fa = f * E2/(E + E2) (Strasburger et al., 2011).

465 iii. Foveal acuity is better for achromatic targets than for achromatic targets, and the

466 proportionality constant E2 is higher (acuity declines slower with eccentricity for achromatic

467 than for chromatic targets) (R. S. Anderson et al., 2002; S. J. Anderson et al., 1991; Noorlander

468 et al., 1983).

469 iv. The high spatial frequency decline in contrast sensitivity at any eccentricity is exponential (Yang

470 et al., 1995): S(f) ~ n-f. This holds over all color channels. (Figure 10)

471 v. Contrast sensitivity for a target of any spatial frequency declines exponentially with eccentricity,

472 with a steeper exponent for higher spatial frequencies (S. J. Anderson et al., 1991; Pointer &

473 Hess, 1989): S(E) ~ n-E. This also holds over all color channels. (Figure 10)

474 vi. The visual system is low-pass: sensitivity across the visual field converges for very low spatial

475 frequencies (Pointer & Hess, 1989).

476 vii. Contrast sensitivity (as d´) for targets of increasing contrast follows an expansive/compressive

477 power function (threshold-vs-contrast functions are dipper-shaped) (G. E. Legge & Foley, 1980;

478 Nachmias & Sansbury, 1974).

479 viii. “Contrast constancy”: In the absence of other interactions, contrast responses converge at

480 high contrasts, for mechanisms tuned to different spatial frequency, orientation, and/or
 Figure 10. Two overarching features of human contrast sensitivity. Contrast sensitivity declines
with retinal eccentricity and with decreasing scale. Black lines show the exponential relation of
contrast sensitivity to eccentricity (left) and spatial frequency (right). An increase in either
parameter corresponds to decreased sensitivity. Dashed red lines show a more accurate model
with a low-frequency plateau – only the lowest spatial frequencies are significantly different.

481 eccentricity (Mark W. Cannon, 1985; C.-C. Chen et al., 2000; Georgeson & Sullivan, 1975;

482 Swanson & Wilson, 1985).

483 ix. Sensitivity for a low-contrast target of one orientation is strongly impaired by a high-contrast

484 overlaid mask of very different orientation (‘cross-orientation masking’), while a high-contrast

485 target is relatively unaffected by a lower-contrast mask (J. M. Foley, 1994).

486 x. The combined perceptual response to contrast over multiple frequency bands is a high-p-norm

487 (less-than optimal combination: M~=4) (M. W. Cannon & Fullenkamp, 1991).

488

489 4.2 Some other (unmodeled) patterns

490 These patterns are less pertinent to the question of color and sharpness percepts across the

491 visual field. Many of these are aggregations of contradictory interactions (i.e. of facilitation and

492 suppression), so we can think of them as ‘washing out’. Others are factors that, by their omission,
493 mean the model is probably underestimating rather than overestimating sensitivity. These are

494 significant aspects of spatial vision psychophysics that touch on the questions at hand, so they

495 warrant acknowledgement:

496 xi. The temporal aspect of contrast sensitivity is ignored. The model’s sensitivity is effectively

497 based on the typical 100-500ms pulse-windowed targets in many traditional trial-based

498 psychophysics tasks (i.e. temporally-broadband and generally low-pass or low-frequency-

499 biased). This can be reasonably linked (as in (Haun & Peli, 2013b)) to the typical duration of

500 fixations between saccades (Henderson, 2003).

501 xii. Orientation anisotropies (oblique effects and horizontal effects (Essock et al., 2009) and radial

502 anisotropies (Westheimer, 2003), etc) of various types are small and ignored here.

503 xiii. Surround-modulation and other lateral-interaction effects (C. C. Chen & Tyler, 2008; Yu et

504 al., 2003) are ignored.

505 xiv. Interactions between the color channels (e.g. Kim & Mullen, 2016) are ignored.

506 xv. Contrast adaptation effects are ignored.

507 xvi. The low-spatial-frequency roll-off in contrast sensitivity has a functional form that I do not

508 attempt to match closely.

509 xvii. Cross-orientation masking is dependent on stimulus parameters and is usually weaker than

510 self(same-orientation)-masking (J. M. Foley, 1994; Meese & Holmes, 2007), but I take it to

511 be of constant strength equivalent to the strength of a filter’s self-masking, i.e. =1. Thus the

512 model always overestimates the strength of cross-orientation masking.

513 xviii. Cross-frequency interactions are captured only by the cross-frequency norm (section 4.7),

514 ignoring specific patterns of interaction (Haun & Essock, 2010; Haun & Peli, 2013b).

515 xix. Contrast sensitivity is known to improve on virtually every dimension when the scene’s

516 luminance is increased, and natural scenes generally have far higher luminances than
517 experimental contrast displays; the current model is fixed to the low-luminance, but still

518 photopic, experimental context (so sensitivity may generally be underestimated by the model).

519 xx. Binocularity adds many complexities to spatial vision. The contrast sensitivity data that the

520 model is fitted to was all collected monocularly; binocular sensitivity and discrimination is

521 generally better than monocular (Campbell & Green, 1965; Gordon E. Legge, 1984), so the

522 estimates of sensitivity can generally be considered underestimates on this point.

523 xxi. In its fine structure, color perception across the visual field has many idiosyncrasies that are

524 left out of the model. For example, in many observers ‘green’ perception is gradually more

525 attenuated towards the periphery (Charles F. Stromeyer et al., 1992), while R/G contrast

526 sensitivity continues to follow the expected pattern. At the fovea, there are no S-cones and

527 so foveal sensitivity to blue spots is very poor (Wald, 1967), although foveal blue percepts still

528 are possible (Magnussen et al., 2001).

529

530 4.3 Threshold over eccentricity and scale

531 Patterns i-x are interrelated in various ways. Of particular importance, ii., iv., and v. (regarding

532 relations of sensitivity with scale and eccentricity) are distinct aspects of the scale-sensitivity of the

533 visual system, and they are modeled compactly:

f E+E2
534 t(f, E) = t 0 exp ((f ) ( E2
)) L(f, E) (7)
0

535 Here L is the low-frequency plateau in sensitivity, which I take to be independent of eccentricity (vi.).
α
536 L(f, E) = 1 + 2 (8)
1+(f⁄f )
1

537 Regarding the high-frequency component of the function (Eq.7, disregarding L), notice how the

538 exponential declines of sensitivity with eccentricity (E) and spatial frequency (f) are intertwined. A

539 single exponential term in the contrast sensitivity function equation captures both features of contrast
540 sensitivity, and their interdependence. There are two constants: f 0 captures the steepness of the

541 decline in sensitivity with increasing spatial frequency (decreasing scale) and E 2 captures the

542 steepness of the decline in sensitivity with increasing eccentricity. In the literature these two

543 parameters are normally considered separately (see (Watson, 2018) for a recent synthesis), but it is

544 clear that they modify each other: if f0 is considered as constant, then E2 is effectively reduced as

545 frequency increases: the decline in sensitivity for some target (f) with eccentricity (E) becomes

546 steeper for higher frequencies (Figure 10A). Conversely, if E2 is considered as constant, then f0 is

547 effectively reduced as eccentricity (E) increases, meaning that the decline in threshold with frequency

548 gets steeper (Figure 10B).

549 This connection between the contrast sensitivity functions for spatial frequency and eccentricity

550 has an interesting consequence for sensitivity across the visual field: it means that sensitivity to some

551 frequency F1 at some eccentricity E1 can usually be matched by sensitivity to a lower frequency F2 at

552 some larger eccentricity E2, or to a higher frequency F0 at some smaller eccentricity E0. The extreme

553 case of this is the acuity frequency: sensitivity to this frequency, which declines with eccentricity, is

554 defined everywhere across the visual field as a constant value (threshold=1). There are, likewise,

555 mechanisms across the visual field with t=½, t=¼, and so-on, with the rule breaking down only at a

556 very low (and to my knowledge, unknown) contrast.

557

558 4.4 Suprathreshold sensitivity

559 The aspects of contrast sensitivity relating to suprathreshold contrasts (vii, viii, and x), and the

560 capacity of the model to reproduce in detail psychophysical task performance (e.g. trialwise behavior

561 and psychometric function shape) are all captured with the Foley transducer (J. M. Foley, 1994; G.

562 E. Legge & Foley, 1980; C. F. Stromeyer & Klein, 1974) that defines perceptual response as a signal-

563 noise ratio (d´) underlying detection and discrimination judgments:

 Figure 11. Model of contrast perception. A) At every point in the model visual field, there is a
complement of filters. The scale of the complement only increases with eccentricity. B) Each filter
complement is divided into three color-opponent channels. C) Each color-opponent channel is
composed of filters tuned to each combination of four scales and four orientations. D) Exposed
to a stimulus, each filter yields a complex output consisting of a contrast magnitude (C) and its
phase. E) Contrast magnitude is transduced to a signal-noise ratio (d´). The nonlinearity depends
on filter scale and eccentricity (Eqs 7,8), and on local contrast at other orientations (Eq 9).

(gcθ )p+q
564 d′θ = R max (9)
z +∑θ (gc)p
p

565 In Eq.3, p, q, and Rmax are fixed over all other parameters (eccentricity, orientation, frequency, color

566 channel). z is the one parameter that varies with eccentricity, frequency, and color. The variable c is

567 the linear response of a spatial filter to the stimulus. The linear gain g of each filter is a partial

568 parameter that ordinarily would be at unity and distributed between R max and z (Haun, 2009); it was

569 considered separately here as it can account in a computationally-convenient way for significant

570 differences in sensitivity between color channels, allowing other parameters to be fixed. The

571 summation in the denominator is over other orientations of same-location, same-frequency, same-

572 color filters. The threshold for each filter is set by transforming the detection threshold functions

573 defined in equations 1 and 2, assuming a fixed experimental d´ value (d´=2 was used in all simulations

574 in this paper):

 𝑅𝑚𝑎𝑥
575 𝑧𝑝 = 𝑡𝜃 𝑝+𝑞 − 𝑡𝜃 𝑝 (10)
𝑑′𝜃

576 Eq. 4 is just a rearrangement of Eq. 3, ignoring the cross-orientation terms (basic contrast sensitivity

577 functions are established without masks) and setting contrast equal to an empirical threshold value.

578

579 2.5 Opponent color channels

580 The three color channels were implemented by transforming input RGB images into the CIElab

581 color space – which consists of a luminance channel (Achr), a blue-yellow channel (B/Y)2, and a red-

582 green channel (R/G) – and using these as inputs to the filter layer. The CIElab components are distinct

583 from the ‘cone contrast’ components typically employed in color psychophysics, but they are close

584 enough for present purposes (McDermott & Webster, 2012). CIElab is widely available in different

585 code bases and is clearly documented in many sources, making the present results more easily

586 replicable.

587

588 2.6 Filter structure

589 The model is framed in a rectangular array of visual field locations, with each location a 2d

590 coordinate in degrees eccentricity from a point of fixation. At each location a complement of filters is

591 assigned: filters tuned to each combination of four (or three) spatial frequencies, four orientations,

592 and three color contrasts. The filters are created on a log-frequency / orientation domain (eq.11). The

593 frequency profile is a Gaussian (eq.12), and the orientation profile is a raised cosine (eq.13):

594 𝑔(𝑓, 𝜃) = 𝑔(𝑓) ∙ 𝑔(𝜃) (11)

2
This implementation of the B/Y channel ignores the fact that human spectral sensitivity is far better for yellow
spots than for blue spots (Abramov & Gordon, 1977). The human visual system is not so simple as to
symmetrically encode blue and yellow, but the model is, so it tends to underestimate yellow and overestimate
blue contributions to stimuli. A similar problem exists for the achromatic channel, in that sensitivity to ‘dark’
spots is generally better than for ‘bright’ spots (Haun & Peli, 2013b; Whittle, 1992). Psychophysically, green
and red are relatively more symmetric, though see point xxi in section 2.2.
 2
(log2 𝑓−log2 𝑓𝑝𝑒𝑎𝑘 )
595 𝑔(𝑓) = exp (− 2𝜎𝑓 2
) (12)

1
(cos[𝜎𝜃 (𝜃 − 𝜃𝑝𝑒𝑎𝑘 )] + 1), 𝑖𝑓 |𝜃 − 𝜃𝑝𝑒𝑎𝑘 | < 𝜎𝜃 𝜋
596 𝑔(𝜃) = {2 (13)
0, 𝑖𝑓 |𝜃 − 𝜃𝑝𝑒𝑎𝑘 | ≥ 𝜎𝜃 𝜋

597

598 Frequency bandwidths were fixed at 2σf =1.4 octaves for all filters (Wilson et al., 1983).

599 Orientation bandwidth was fixed at 45 degrees (σθ = ¼π). In the space domain, these are quadrature

600 filters, meaning they encode both amplitude and phase of content in their pass band; this is equivalent

601 to each filter’s being a set of four half-rectified filters, two with opposite sine phases and two with

602 opposite cosine phases. The amplitude of each filter was adjusted so that the linear response (dot-

603 product) of the filter to a full-contrast sinewave grating pattern at its preferred orientation and

604 frequency would be unity, i.e. the linear response of each filter was explicitly equated with the

605 ‘contrast energy’ (the phase-invariant Michelson contrast magnitude) of a typical contrast sensitivity

606 test stimulus.

607 The four peak orientations of the filters, at every visual field location and for each color channel,

608 were [0, 45, 90, 135] from vertical. The peak frequencies were dependent on both eccentricity and

609 color channel. The scheme was as follows: first, the contrast sensitivity function of Eq(1) was fit to

610 data sets for human sensitivity to achromatic (Pointer & Hess, 1989), red-green (S. J. Anderson et

611 al., 1991), and blue-yellow (K. T. Mullen & Kingdom, 2002) contrast stimuli of variable spatial

612 frequency and eccentricity. The ‘local scale’ of the system’s filter complements (Watson, 1987) was

613 set according to the objective acuity at that eccentricity. The acuity frequency for each channel as a

614 function of eccentricity was derived by setting threshold t(f,E) = 1 and solving for spatial frequency f.

615 The peak frequencies (fpeak) of the filters for each channel at each eccentricity were set to:

𝑓(𝑡=1,𝐸)
616 {𝑓𝑝𝑒𝑎𝑘 (𝐸 )} = {3,6,12,24} (14)
617 That is, the peak frequency of the finest filter at each location was set to 1/3 of the acuity limit

618 expected by the pattern of contrast sensitivity, and the coarser filters were spaced 1, 2, and 3 octaves

619 below the finest filter. The low-pass residual from these filters was then inserted to capture the low-
 Figure 12. The contrast sensitivity and acuity of the model are closely matched to that of human
observers. The upper panels show human and model contrast sensitivity estimates for
achromatic (left) and red/green (right) targets, for different target spatial frequencies (cycles per
degree, cpd) and retinal eccentricities (in degrees of visual angle, dva). The straight lines are
regression fits to the log-CS human data, and are the same in both data and model plots, to aid
in by-eye comparison. The lower panels show acuity for achromatic, blue/yellow, and red/green
grating stimuli; solid symbols are human acuity, open symbols are model acuity. The solid lines
are linear fits to the human data, and are the same in both human and model panels.

620 frequency plateau of the sensitivity function. This close packing of the frequency bands allowed the

621 filter complement to scale flexibly with eccentricity, avoiding most cases where the coarsest filters

622 might be too large for the test image. In cases where the lowest-frequency filter size approached the

623 size of the test image (i.e. chromatic filters at large eccentricities), 3 rather than 4 filter scales were

624 used.

625 Each individual filter yields a ‘c’ value in response to the stimulus image. The signal-noise ratio

626 (d´(c)) for each filter is derived according to Equation (9). This matrix of d´ values can then be

627 analyzed as a detailed model of perceptual response to contrast.

628

629 4.7 Fitting model to data

630 The ‘fitting’ stage of model construction was abstract: data was subjected to a curve-fitting

631 procedure to set the free parameters of the CSF model (Eq.7), with starting parameters suggested

632 by the papers where the model components were found (Geisler & Perry, 1998; Watson & Ahumada,

633 2005; Yang et al., 1995). The nonlinear contrast response parameters (Rmax, p, q) were fixed at the

634 outset to standard values (Haun & Peli, 2013b; G. E. Legge & Foley, 1980). The target data were

635 drawn from the studies listed in Table 1. I did not directly fit B/Y contrast sensitivity data, and instead

636 scaled the high-s.f. decline constant (f0) of the achromatic sensitivity function (see Table 2), and

637 decreased the linear gain, to obtain a good fit to the B/Y acuity data.

638
 Data type Source

Achromatic CS by spatial (Pointer & Hess, Figures 2,

frequency and eccentricity 1989) 3, 12
Red/Green CS by spatial (K. T. Mullen, Figures 3,
frequency and eccentricity 1991) 4
Achromatic and Red/Green (S. J. Anderson et
Figure 5
acuity by eccentricity al., 1991)
Achromatic and Blue/Yellow (R. S. Anderson
Figure 3
acuity by eccentricity et al., 2002)
639 Table 1.

640

641 To reproduce the psychophysics of the combination of spatial overlapping responses, we sum

1/𝑀
642 responses across scale with a high p-norm 𝑅̅ = (∑𝑓 𝑅(𝑓)𝑀 ) , with M=4; (M. W. Cannon &

643 Fullenkamp, 1991; Haun & Peli, 2013b)), and then take the maximum response across the remaining

644 filter dimensions. This is the ‘cross-frequency response’ at each location, and it represents the signal-

645 noise ratio the observer has for making decisions in a SDT task. I constructed achromatic and color

646 Gabor patch stimuli of varying spatial frequency and center eccentricity, to (roughly) match the

647 parameters of the exemplar experiments. The contrast of each stimulus was set to produce a peak

648 cross-frequency response equal to the ‘experimental’ d´=2 (an unbiased yes/no hit-rate of 84%) –

649 this contrast was the model’s threshold for the target stimulus.

650 The model’s contrast thresholds are plotted in Figure 12 against the data. Also shown are acuity

651 estimates – these are obtained by extrapolating the contrast sensitivity curves to the spatial frequency

652 where sensitivity = 1. The model generally comes very close to human performance, close enough

653 that we can consider the following claim to be reasonable: whatever conclusions one might draw

654 from the empirical data should be transferrable to the simulated data. That is, if one believes (for

655 example) that human color or sharpness representations must decline in quality due to the observed

656 pattern of contrast sensitivity, then the same belief should apply to the representations of this model.
657 Put another way, the facts pertinent to claims about what a human observer can and cannot

658 experience at the level of contrast perception are all effectively embodied by this model.

659 In addition to these basic contours of contrast sensitivity, the model also captures – more

660 qualitatively – some further aspects of spatial vision. Contrast discrimination (threshold-vs-contrast)

661 functions have the familiar ‘dipper’ shape, where increment thresholds for just-detectable pedestals

662 are smaller than the detection threshold itself, while increment thresholds for suprathreshold

663 pedestals follow a compressive power function (Stevens’ law). Cross-oriented masks elevate the

664 dipper functions in the expected way, though they are much more powerful than they should be (since

665 the cross-oriented inputs were not realistically weighted).

666

667

668

669

670

671
672

Value Description

Achr 0.0051

t0 B/Y 0.0051 Overall sensitivity function amplitude (adjusted by g)

R/G 0.0082

Achr 6.22

E2 B/Y 6.22 Eccentricity-scaling constant

R/G 1.82

Achr 5.26

α B/Y 5.26 Low-s.f. threshold weight

R/G 2.53

Achr 4.51

f0 B/Y 4.51 / 6 High-s.f. decline constants

R/G 5.37

Achr 0.32

f1 B/Y 0.32 Low-s.f. decline constant

R/G 0.94

Achr 2.5

g B/Y 1 Linear (filter) gain

R/G 2

Rmax 30 Nonlinear gain

p 2 Low-c response exponent

q 0.4 High-c response exponent

673 Table 2

674

675
676 Acknowledgments

677 Thanks to Yeon Jin Kim for helpful discussion on color vision psychophysics, and Giulio Tononi for

678 support and comments on the manuscript. This work was stimulated by discussion of Haun et al 2017

679 published at the Brains Blog (Burnston 2018), and was presented at the 2019 Association for

680 Scientific Study of Consciousness meeting in London, Ontario. Andrew Haun is supported by a grant

681 from the Tiny Blue Dot Foundation to the UW-Madison Center for Sleep and Consciousness.

682

683 5. References

684 Abramov, I., & Gordon, J. (1977). Color vision in the peripheral retina. I. Spectral sensitivity. JOSA,

685 67(2), 195–202.

686 Adelson, E. H. (2000). Lightness Perception and Lightness Illusions. In The New Cognitive

687 Neurosciences (2nd ed., pp. 339–351). MIT Press, Cambridge, MA.

688 Anderson, R. S., Zlatkova, M. B., & Demirel, S. (2002). What limits detection and resolution of short-

689 wavelength sinusoidal gratings across the retina? Vision Research, 42(8), 981–990.

690 https://doi.org/10.1016/S0042-6989(02)00013-5

691 Anderson, S. J., Mullen, K. T., & Hess, R. F. (1991). Human peripheral spatial resolution for achromatic

692 and chromatic stimuli: Limits imposed by optical and retinal factors. The Journal of Physiology,

693 442(1), 47–64.

694 Anstis, S. (1998). Picturing peripheral acuity. PERCEPTION-LONDON-, 27, 817–826.

695 Balas, B., & Sinha, P. (2007). “Filling-in” colour in natural scenes. Visual Cognition, 15(7), 765–778.

696 Block, N. (2001). Paradox and cross purposes in recent work on consciousness. Cognition, 79(1–2),

697 197–219.

698 Block, N. (2007). Consciousness, accessibility, and the mesh between psychology and neuroscience.

699 Behavioral and Brain Sciences, 30(5–6), 481–499; discussion 499-548.

700 http://dx.doi.org/10.1017/S0140525X07002786
701 Bressler, D. W., Fortenbaugh, F. C., Robertson, L. C., & Silver, M. A. (2013). Visual spatial attention

702 enhances the amplitude of positive and negative fMRI responses to visual stimulation in an

703 eccentricity-dependent manner. Vision Research, 85, 104–112.

704 https://doi.org/10.1016/j.visres.2013.03.009

705 Buchsbaum, G., & Gottschalk, A. (1983). Trichromacy, opponent colours coding and optimum colour

706 information transmission in the retina. Proceedings of the Royal Society of London. Series B.

707 Biological Sciences, 220(1218), 89–113.

708 Buracas, G. T., & Boynton, G. M. (2007). The effect of spatial attention on contrast response functions

709 in human visual cortex. Journal of Neuroscience, 27(1), 93–97.

710 Burnston, D. (2018, April 13). Symposium on Haun, Tononi, Koch, and Tsuchiya: “Are we

711 underestimating the richness of visual experience?” The Brains Blog.

712 http://philosophyofbrains.com/2018/04/13/symposium-on-haun-tononi-koch-and-tsuchiya-are-

713 we-underestimating-the-richness-of-visual-experience.aspx

714 Burton, G. J., & Moorhead, I. R. (1987). Color and spatial structure in natural scenes. Applied Optics,

715 26(1), 157–170.

716 Campbell, F. W., & Green, D. G. (1965). Monocular versus binocular visual acuity. Nature, 208(5006),

717 191–192.

718 Cannon, M. W., & Fullenkamp, S. C. (1991). A Transducer Model for Contrast Perception. Vision

719 Research, 31, 983–998. https://doi.org/10.1016/s0042-6989(05)80001-x

720 Cannon, Mark W. (1985). Perceived contrast in the fovea and periphery. JOSA A, 2(10), 1760–1768.

721 Carrasco, M. (2011). Visual attention: The past 25 years. Vision Research, 51(13), 1484–1525.

722 https://doi.org/10.1016/j.visres.2011.04.012

723 Carrasco, M., & Barbot, A. (2019). Spatial attention alters visual appearance. Current Opinion in

724 Psychology, 29, 56–64. https://doi.org/10.1016/j.copsyc.2018.10.010

725 Carrasco, M., Ling, S., & Read, S. (2004). Attention alters appearance. Nature Neuroscience, 7(3),

726 308–313. https://doi.org/10.1038/nn1194

727 Carrasco, M., Penpeci-Talgar, C., & Eckstein, M. (2000). Spatial covert attention increases contrast

728 sensitivity across the CSF: Support for signal enhancement. Vision Research, 40(10–12), 1203–

729 1215.

730 Chen, C. C., & Tyler, C. W. (2008). Excitatory and inhibitory interaction fields of flankers revealed by

731 contrast-masking functions. Journal of Vision, 8, 14. https://doi.org/10.1167/8.4.10

732 Chen, C.-C., Foley, J. M., & Brainard, D. H. (2000). Detection of chromoluminance patterns on

733 chromoluminance pedestals I: Threshold measurements. Vision Research, 40(7), 773–788.

734 https://doi.org/10.1016/S0042-6989(99)00227-8

735 Cohen, M. A., Dennett, D. C., & Kanwisher, N. (2016). What is the Bandwidth of Perceptual Experience?

736 Trends in Cognitive Sciences, 20(5), 324–335. https://doi.org/10.1016/j.tics.2016.03.006

737 Curcio, C. A., Sloan, K. R., Kalina, R. E., & Hendrickson, A. E. (1990). Human photoreceptor

738 topography. Journal of Comparative Neurology, 292(4), 497–523.

739 Dennett, D. C. (2005). Sweet dreams: Philosophical obstacles to a science of consciousness. MIT

740 press.

741 Dobelle, W. H., & Mladejovsky, M. G. (1974). Phosphenes produced by electrical stimulation of human

742 occipital cortex, and their application to the development of a prosthesis for the blind. The

743 Journal of Physiology, 243(2), 553–576. https://doi.org/10.1113/jphysiol.1974.sp010766

744 Elder, J. H., & Zucker, S. W. (1996). Scale space localization, blur, and contour-based image coding.

745 Proceedings CVPR IEEE Computer Society Conference on Computer Vision and Pattern

746 Recognition, 27–34.

747 Elliott, S. L., Georgeson, M. A., & Webster, M. A. (2011). Response normalization and blur adaptation:

748 Data and multi-scale model. Journal of Vision, 11, 18. https://doi.org/10.1167/11.2.7

749 Essock, E. A., Haun, A. M., & Kim, Y. J. (2009). An anisotropy of orientation-tuned suppression that

750 matches the anisotropy of typical natural scenes. Journal of Vision, 9(1).

751 http://www.journalofvision.org/content/9/1/35.short
752 Foley, J. M. (1994). Human Luminance Pattern-Vision Mechanisms—Masking Experiments Require a

753 New Model. Journal of the Optical Society of America A-Optics Image Science and Vision, 11,

754 1710–1719. https://doi.org/10.1364/josaa.11.001710

755 Foley, John M., & Schwarz, W. (1998). Spatial attention: Effect of position uncertainty and number of

756 distractor patterns on the threshold-versus-contrast function for contrast discrimination. JOSA

757 A, 15(5), 1036–1047.

758 Galvin, S. J., O’Shea, R. P., Squire, A. M., & Govan, D. G. (1997). Sharpness overconstancy in

759 peripheral vision. Vision Research, 37(15), 2035–2039. https://doi.org/10.1016/S0042-

760 6989(97)00016-3

761 Galvin, S. J., O’Shea, R. P., Squire, A. M., & Hailstone, D. S. (1999). Sharpness overconstancy: The

762 roles of visibility and current context. Vision Research, 39(16), 2649–2657.

763 https://doi.org/10.1016/S0042-6989(98)00306-X

764 Gandhi, S. P., Heeger, D. J., & Boynton, G. M. (1999). Spatial attention affects brain activity in human

765 primary visual cortex. Proceedings of the National Academy of Sciences, 96(6), 3314–3319.

766 Gegenfurtner, K. R., Bloj, M., & Toscani, M. (2015). The many colours of ‘the dress.’ Current Biology,

767 25(13), R543–R544.

768 Geisler, W. S., & Perry, J. S. (1998). Real-time foveated multiresolution system for low-bandwidth video

769 communication. Proc. SPIE, 3299, 294–305. http://link.aip.org/link/?PSI/3299/294/1

770 http://dx.doi.org/10.1117/12.320120

771 Georgeson, M. A., May, K. A., Freeman, T. C. A., & Hesse, G. S. (2007). From filters to features: Scale

772 space analysis of edge and blur coding in human vision. Journal of Vision, 7.

773 https://doi.org/10.1167/7.13.7

774 Georgeson, M. A., & Sullivan, G. D. (1975). Contrast Constancy—Deblurring in Human Vision by Spatial

775 Frequency Channels. Journal of Physiology-London, 252, 627-.

776 Grusser, O. J. (1995). Migraine Phosphenes and the Retino-Cortical Magnification Factor. Vision

777 Research, 35, 1125–1134. https://doi.org/10.1016/0042-6989(94)00187-q

778 Guo, F., Liu, C., Qian, C., Zhang, Z., Sun, K., Wang, D. J., He, S., & Zhang, P. (2020). Layer-dependent

779 multiplicative effects of spatial attention on contrast responses in human early visual cortex.

780 BioRxiv, 2020.02.01.926303. https://doi.org/10.1101/2020.02.01.926303

781 Haun, A. M. (2009). Contrast Sensitivity in 1/f Noise [University of Louisville].

782 https://doi.org/10.13140/rg.2.1.3905.5528

783 Haun, A. M., & Essock, E. A. (2010). Contrast sensitivity for oriented patterns in 1/f noise: Contrast

784 response and the horizontal effect. Journal of Vision, 10, 21. https://doi.org/10.1167/10.10.1

785 Haun, A. M., & Hansen, B. C. (2019). Visual evoked potentials elicited by complex scenes are regulated

786 by high spatial frequency content. Journal of Vision, 19(10), 123b–123b.

787 Haun, A. M., & Peli, E. (2013a). Adaptation to blurred and sharpened video. Journal of Vision, 13(8).

788 http://ww.journalofvision.org/content/13/8/12.short

789 Haun, A. M., & Peli, E. (2013b). Perceived contrast in complex images. Journal of Vision, 13(13), 3–3.

790 https://doi.org/10.1167/13.13.3

791 Haun, A. M., Tononi, G., Koch, C., & Tsuchiya, N. (2017). Are we underestimating the richness of visual

792 experience? Neuroscience of Consciousness, 2017(1). https://doi.org/10.1093/nc/niw023

793 Henderson, J. M. (2003). Human gaze control during real-world scene perception. Trends in Cognitive

794 Sciences, 7(11), 498–504.

795 Herrmann, K., Kouhsari, L. M., Carrasco, M., & Heeger, D. J. (2009). Behavioral studies support the

796 normalization model of attention. Perception, 38, 6–6.

797 Hess, R. F., Pointer, J. S., & Watt, R. J. (1989). How are spatial filters used in fovea and parafovea?

798 JOSA A, 6(2), 329–339.

799 Huang, L., & Dobkins, K. R. (2005). Attentional effects on contrast discrimination in humans: Evidence

800 for both contrast gain and response gain. Vision Research, 45(9), 1201–1212.

801 Intriligator, J., & Cavanagh, P. (2001). The spatial resolution of visual attention. Cognitive Psychology,

802 43(3), 171–216.

803 Kahn, E., Dement, W., Fisher, C., & Barmack, J. E. (1962). Incidence of Color in Immediately Recalled

804 Dreams. Science, 137(3535), 1054–1055. https://doi.org/10.1126/science.137.3535.1054

805 Kayargadde, V., & Martens, J.-B. (1996). Perceptual characterization of images degraded by blur and

806 noise: Experiments. JOSA A, 13(6), 1166–1177. https://doi.org/10.1364/JOSAA.13.001166

807 Kim, Y. J., & Mullen, K. T. (2016). Effect of overlaid luminance contrast on perceived color contrast:

808 Shadows enhance, borders suppress. Journal of Vision, 16(11), 15–15.

809 https://doi.org/10.1167/16.11.15

810 Koenderink, J. J. (1984). The structure of images. Biological Cybernetics, 50(5), 363–370.

811 Krauskopf, J., Williams, D. R., & Heeley, D. W. (1982). Cardinal directions of color space. Vision

812 Research, 22(9), 1123–1131. https://doi.org/10.1016/0042-6989(82)90077-3

813 Kurihara, T., Aoki, N., & Kobayashi, H. (2009). Analysis of sharpness increase by image noise. Proc.

814 SPIE, 724014.

815 Lau, H., & Brown, R. (2019). The emperor’s new phenomenology? The empirical case for conscious

816 experiences without first-order representations. In Blockheads!: Essays on Ned Block’s

817 Philosophy of Mind and Consciousness (pp. 171–197).

818 Lee, D. K., Koch, C., & Braun, J. (1997). Spatial vision thresholds in the near absence of attention.

819 Vision Research, 37(17), 2409–2418.

820 Legge, G. E., & Foley, J. M. (1980). Contrast Masking in Human-Vision. Journal of the Optical Society of

821 America, 70, 1458–1471. https://doi.org/10.1364/josa.70.001458

822 Legge, Gordon E. (1984). Binocular contrast summation—I. Detection and discrimination. Vision

823 Research, 24(4), 373–383.

824 Levi, D. M., & Klein, S. A. (1990). Equivalent intrinsic blur in spatial vision. Vision Research, 30(12),

825 1971–1993. https://doi.org/10.1016/0042-6989(90)90016-E

826 Lu, Z.-L., & Dosher, B. A. (1999). Characterizing human perceptual inefficiencies with equivalent

827 internal noise. JOSA A, 16(3), 764–778.

828 Luck, S. J., Chelazzi, L., Hillyard, S. A., & Desimone, R. (1997). Neural Mechanisms of Spatial Selective

829 Attention in Areas V1, V2, and V4 of Macaque Visual Cortex. Journal of Neurophysiology, 77(1),

830 24–42. https://doi.org/10.1152/jn.1997.77.1.24

831 Magnussen, S., Spillmann, L., Stürzel, F., & Werner, J. S. (2001). Filling-in of the foveal blue scotoma.

832 Vision Research, 41(23), 2961–2967. https://doi.org/10.1016/S0042-6989(01)00178-X

833 Maiello, G., Walker, L., Bex, P. J., & Vera-Diaz, F. A. (2017). Blur perception throughout the visual field

834 in myopia and emmetropia. Journal of Vision, 17(5), 3–3.

835 McCollough, C. (1965). Color Adaptation of Edge-Detectors in the Human Visual System. Science,

836 149(3688), 1115–1116. https://doi.org/10.1126/science.149.3688.1115

837 McDermott, K. C., & Webster, M. A. (2012). Uniform color spaces and natural image statistics. Journal

838 of the Optical Society of America. A, Optics, Image Science, and Vision, 29(2), A182–A187.

839 Meese, T. S., & Holmes, D. J. (2007). Spatial and temporal dependencies of cross-orientation

840 suppression in human vision. Proceedings of the Royal Society B-Biological Sciences, 274, 127–

841 136. https://doi.org/10.1098/rspb.2006.3697

842 Mullen, K. T. (1991). Colour vision as a post-receptoral specialization of the central visual field. Vision

843 Research, 31(1), 119–130. https://doi.org/10.1016/0042-6989(91)90079-K

844 Mullen, K. T., & Kingdom, F. A. A. (2002). Differential distributions of red-green and blue-yellow cone

845 opponency across the visual field. Visual Neuroscience, 19, 109–118.

846 https://doi.org/10.1017/s0952523802191103

847 Mullen, Kathy T., & Sankeralli, M. J. (1999). Evidence for the stochastic independence of the blue-

848 yellow, red-green and luminance detection mechanisms revealed by subthreshold summation.

849 Vision Research, 39(4), 733–745. https://doi.org/10.1016/S0042-6989(98)00137-0

850 Nachmias, J., & Sansbury, R. V. (1974). Grating Contrast—Discrimination May Be Better Than

851 Detection. Vision Research, 14, 1039–1042. https://doi.org/10.1016/0042-6989(74)90175-8

852 Nebel, B. R. (1957). The phosphene of quick eye motion. AMA Archives of Ophthalmology, 58(2), 235–

853 243.
854 Noorlander, C., Koenderink, J. J., Den Olden, R. J., & Edens, B. W. (1983). Sensitivity to spatiotemporal

855 colour contrast in the peripheral visual field. Vision Research, 23(1), 1–11.

856 Otten, M., Pinto, Y., Paffen, C. L., Seth, A. K., & Kanai, R. (2017). The uniformity illusion: Central stimuli

857 can determine peripheral perception. Psychological Science, 28(1), 56–68.

858 Peli, E. (1996). Test of a model of foveal vision by using simulations. JOSA A, 13(6), 1131–1138.

859 https://doi.org/10.1364/JOSAA.13.001131

860 Perona, P., & Malik, J. (1990). Scale-space and edge detection using anisotropic diffusion. IEEE

861 Transactions on Pattern Analysis and Machine Intelligence, 12(7), 629–639.

862 Pointer, J. S., & Hess, R. F. (1989). The Contrast Sensitivity Gradient across the Human Visual-Field—

863 With Emphasis on the Low Spatial-Frequency Range. Vision Research, 29, 1133-.

864 https://doi.org/10.1016/0042-6989(89)90061-8

865 Portilla, J., & Simoncelli, E. P. (2000). A Parametric Texture Model Based on Joint Statistics of Complex

866 Wavelet Coefficients. International Journal of Computer Vision, 40(1), 49–70.

867 https://doi.org/10.1023/A:1026553619983

868 Rechtschaffen, A., & Buchignani, C. (1992). The Visual Appearance of Dreams. The Neuropsychology

869 of Sleep and Dreaming, 143.

870 Reynolds, J. H., & Heeger, D. J. (2009). The normalization model of attention. Neuron, 61(2), 168.

871 Richards, W. (1971). The fortification illusions of migraines. Scientific American.

872 Robson, J. G., & Graham, N. (1981). Probability summation and regional variation in contrast sensitivity

873 across the visual field. Vision Research, 21(3), 409–418. https://doi.org/10.1016/0042-

874 6989(81)90169-3

875 Rohatgi, A. (2017). WebPlotDigitizer. Austin, Texas, USA.

876 Rosenholtz, R. (2016). Capabilities and Limitations of Peripheral Vision. Annual Review of Vision

877 Science, 2(1), 437–457. https://doi.org/10.1146/annurev-vision-082114-035733

878 Ruderman, D. L., & Bialek, W. (1994). Statistics of natural images: Scaling in the woods. Physical

879 Review Letters, 73(6), 814–817.

880 Schneider, K. A., & Komlos, M. (2008). Attention biases decisions but does not alter appearance.

881 Journal of Vision, 8(15), 3–3. https://doi.org/10.1167/8.15.3

882 Schwitzgebel, E. (2008). The Unreliability of Naive Introspection. Philosophical Review, 117(2), 245–

883 273. https://doi.org/10.1215/00318108-2007-037

884 Smith, A. R. (1978). Color gamut transform pairs. ACM Siggraph Computer Graphics, 12(3), 12–19.

885 Strasburger, H., Rentschler, I., & Juttner, M. (2011). Peripheral vision and pattern recognition: A review.

886 Journal of Vision, 11. https://doi.org/10.1167/11.5.13

887 Stromeyer, C. F., & Klein, S. (1974). Spatial frequency channels in human vision as asymmetric (edge)

888 mechanisms. Vision Research, 14(12), 1409–1420.

889 Stromeyer, Charles F., Lee, J., & Eskew Jr, R. T. (1992). Peripheral chromatic sensitivity for flashes: A

890 post-peceptoral red-green asymmetry. Vision Research, 32(10), 1865–1873.

891 Swanson, W. H., & Wilson, H. R. (1985). Eccentricity dependence of contrast matching and oblique

892 masking. Vision Research, 25(9), 1285–1295.

893 Tyler, C. W. (1978). Some New Entoptic Phenomena. Vision Research, 18, 1633–1639.

894 https://doi.org/10.1016/0042-6989(78)90255-9

895 Tyler, Christopher W. (2015). Peripheral Color Demo. I-Perception, 6(6), 2041669515613671.

896 Van Gulick, R. (2007). What if phenomenal consciousness admits of degrees? Behavioral and Brain

897 Sciences, 30(5–6), 528–529.

898 (Various). (2019, Summer). Https://en.wikipedia.org/wiki/Wikipedia:Featured_pictures/Places/Panorama.

899 Wikipedia.

900 Wald, G. (1967). Blue-blindness in the normal fovea. JOSA, 57(11), 1289–1301.

901 Wang, B., & Ciuffreda, K. J. (2005). Blur discrimination of the human eye in the near retinal periphery.

902 Optometry and Vision Science, 82(1), 52–58.

903 Wang, Z., & Simoncelli, E. P. (2003). Local phase coherence and the perception of blur. Adv. Neural

904 Information Processing Systems (NIPS03), 16.

905 http://books.nips.cc/papers/files/nips16/NIPS2003_VB01.pdf
906 Watson, A. B. (1987). Estimation of Local Spatial Scale. Journal of the Optical Society of America A-

907 Optics Image Science and Vision, 4, 1579–1582. https://doi.org/10.1364/josaa.4.001579

908 Watson, A. B. (2018). The Field of View, the Field of Resolution, and the Field of Contrast Sensitivity.

909 Electronic Imaging, 2018(14), 1–11.

910 Watson, A. B., & Ahumada, A. J. (2005). A standard model for foveal detection of spatial contrast.

911 Journal of Vision, 5, 717–740. https://doi.org/10.1167/5.9.6

912 Watson, A. B., & Ahumada, A. J. (2016). The pyramid of visibility. Electronic Imaging, 2016(16), 1–6.

913 Webster, M. A., Georgeson, M. A., & Webster, S. M. (2002). Neural adjustments to image blur. Nature

914 Neuroscience, 5, 839–840. https://doi.org/10.1038/nn906

915 Westheimer, G. (2003). Meridional anisotropy in visual processing: Implications for the neural site of the

916 oblique effect. Vision Research, 43(22), 2281–2289.

917 Whittle, P. (1992). Brightness, discriminability and the “crispening effect.” Vision Research, 32(8), 1493–

918 1507.

919 Wilson, H. R., McFarlane, D. K., & Phillips, G. C. (1983). Spatial-Frequency Tuning of Orientation

920 Selective Units Estimated by Oblique Masking. Vision Research, 23, 873–882.

921 https://doi.org/10.1016/0042-6989(83)90055-x

922 Witkin, A. P. (1987). Scale-space filtering. In Readings in Computer Vision (pp. 329–332). Elsevier.

923 Yang, J., Qi, X. F., & Makous, W. (1995). Zero-Frequency Masking and a Model of Contrast Sensitivity.

924 Vision Research, 35, 1965–1978. https://doi.org/10.1016/0042-6989(94)00285-t

925 Yu, C., Klein, S. A., & Levi, D. M. (2003). Cross- and Iso-oriented surrounds modulate the contrast

926 response function: The effect of surround contrast. Journal of Vision, 3, 527–540.

927 https://doi.org/10.1167/3.8.1

928