The computational neuroscience of animal camouflage

From the moment 'The Origin of Species' was published in 1859, Darwin and his contemporaries used animal camouflage to illustrate how improved survival leads to apparent design. The ideas of early evolutionary biologists (and it is probably significant that many were artists too) in turn had a strong influence on the adoption of camouflage by the military in both World Wars. The basic principles, advanced then,? of blending into the background, use of disruptive patterns to disguise shape and form, and mimicking of background objects -? have remained largely unchanged, textbooks in biology and the visual sciences illustrated by beguiling photos of leaf-like moths, twig-like caterpillars and oddly striped frogs. Surprisingly, however, many of the fundamental principles of camouflage remain untested and, importantly, are stated in language that predates modern computational theories of vision. Our aim is to use modern computational theories of vision to understand the design of animal camouflage, as seen by both humans and animal predators. The great advantage of a computational approach is that it is explicit about the mechanisms of perception and cognition of the human or animal viewer, and so sufficiently precise to generate testable predictions. Of great significance to biologists too, such models can be adjusted to cater for animals with different visual systems from our own. If we are interested in why a moth has a particular colour pattern, we need to understand how its bird predators would see that pattern, and birds not only have a richer colour world than humans (they can see ultraviolet light, for example) their acuity and sensitivity to contrast differs from us. We focus on the two major forms of camouflage, background matching or blending, and disruptive coloration, using computer models and experiments on humans in the lab and wild birds in the field. Bringing the computational neuroscience of vision to biology has clear benefits, but the flow of ideas is not one-way. Because visual systems have evolved to solve real-world problems, of which 'camouflage breaking' is one, then many design features of human vision should be explicable with reference to the ecology of early humans and other primates. Just as we seek to modernise the biological study of coloration through infusion of the theory and technology of computational neuroscience, so too we wish to free the latter of the (usually unrecognised) constraints of modelling the world through human eyes.

We will develop a computational theory of animal camouflage, with models specific to the visual systems of birds and humans. Birds, because many spectacular examples of camouflage come from insects, and so we must understand their coloration with respect to the vision of their major predators. Humans, because of the fundamental interest, from psychology and computer science, in visual search. Previous research has highlighted key differences in the visual systems of birds and humans, but search for cryptic targets against complex natural backgrounds is a task which both visual systems have surely evolved to perform efficiently. Assessing whether the different designs of the bird and primate eye, and neural architecture, have favoured different solutions to this common problem is of fundamental interest. We focus on the two major forms of camouflage, background matching and disruptive coloration. The former will be addressed using a model of spatiochromatic differences to quantify the conspicuousness of cryptic targets. The predictions will be tested with experiments on humans and wild birds in the field and lab. Disruptive coloration can be distinguished from background matching because it acts against object recognition rather than detection. We propose that several mechanisms may be exploited in various types of coloration. First, the outline of the animal may be disguised by high contrast patches at the body's edge, the salience to edge detectors in the predator's visual system creating false contours that break up the shape of the animal. The second way in which disruptive patterns may be beneficial is in disguising otherwise conspicuous body parts, such as eyes and limbs. Finally, there may be disruptive effects even when patches aren't at the body edge, through untested effects of 'visual crowding' or distraction of attention. Disruptive coloration will be investigated through a different class of model, namely object recognition models from Computer Vision.