I managed two impossible things before lunch today[1]. I crossed the road this morning, confident I could avoid the big red bus that I saw in my peripheral vision. I read the links in small-type in the Word Press sidebar, despite the fact that they are in blue text. However, if my retina is to be believed, this shouldn’t be possible.

Colour photo-receptors (cones) are not distributed evenly across the retina. In the fovea, “red” and “green” cones are densely packed, and there are no “blue” cones and no rods. So detailed vision takes place with no (direct) knowledge of short wave light.

However, only 5° of visual angle from the fovea the density of the cones drops dramatically (from around 150,000/mm² at the central fovea to <10,000 per mm²) and the density of the rods rises from none at the fovea to >150,000/mm° at around 15° from the centre of the fovea.

5° of visual angle is about two thumbs’ width when held at arms’ length. Hold up your hands with your thumbs side by side(at arms’ length). Keep your hands forward, so they look something like the “bird” hand shadow. Now fixate on your thumbs. Everything around your thumbs is seen by this area of the retina that is much richer in rods than in cones, whereas the thumbs themselves are seen with the part of the retina that is pretty much all cones. Yet the skin on the back of your hand looks just as colourful as the skin on your thumbs.

So why are colours in the periphery and colours in the centre of the visual field perceived so similarly? The short answer is “because it’s useful”. If the things you saw kept changing colour as they moved across your visual field that would make object identification very difficult indeed. But this doesn’t tell us how the visual system “reconstructs” colour. Is the colour “spread” through the visual field by taking the statistics of the centre and applying them to the surround? Or is there are “top-down” effect, such that knowledge of what is in the visual field tells us how it should look.

Balas and Sinha devised a neat experiment to discriminate between these hypotheses. They decided to address the following questions:

  1. “Do observers complete the colour content of natural scenes when larger regions of the image have had colour artificially removed?”
  2. “If colour completion occurs, does it do so more readily from the centre of an image outwards as opposed from the periphery inwards?”
  3. “If colour completion occurs, does it depend on natural scene statistics?”

To answer the first and second questions they created “Colour Chimeras” – images that were desaturated (“greyed out”) either in the centre or at the edges. Volunteers were presented with images that were entirely grey, entirely coloured (“pan-field” coloured), colour centre chimeras or grey centre chimeras. They found that subjects were much more likely to mistake chimeric images for “pan field” colour images than they were to mistake them for grey images. Importantly, if didn’t matter whether the chimera was greyed out at the centre or the edge: the volunteers still saw a significant proportion of the images as pan-field coloured.

To answer the third question the researchers altered the textural and the colour information in the images. In the first experiment the chimeric and non-chimeric images were all natural scenes (beaches, trees etc.) In the second experiment, some volunteers were presented with natural scenes. Some volunteers were presented with scenes in which the colour had been altered to consist of a single hue, so that the colours were not natural, but shapes within the image were still recognisable. Others were presented with scenes in which the textures were changed, so that the structure of objects was no longer recognisable, but the distribution of colours was the same as the original image. Some subjects were presented with images that had both colout and texture manipulations, in which the original objects and colours could no longer be recognised. How would this affect the volunteers’ ability to spot chimeras?

Volunteers were less likely to “fill in” colour when they were presented with the manipulated chimeras instead of the natural scenes. Textural changes reduced the ability to “spread” colour to the rest of the scene, and colour manipulations reduced this ability even more. However when images were manipulated both for colour and texture, subjects were very good at spotting chimeras (or very bad at filling in colour).

The authors conclude that colour spreading is a common perceptual phenomenon (much more common than the occurrence of “grey spreading” – the mis-identification of chimeras as grey images). Furthermore, they conclude that scene statistics provide important perceptual cues that support this colour spreading. So the next time you see a bus in your peripheral fieldand you know that it’s red, it’s probably because you’ve seen red buses before, and not because your retina tells you so.

[1] I know it should be six before breakfast, but it’s very hot today. Vaughan at Mind Hacks is obviously made of sterner stuff.

Balas, B., Sinha, P. (2007). “Filling-in” colour in natural scenes. Visual Cognition, 15 (7), 765-778. DOI: 10.1080/13506280701295453