Concealing 3D objects

Camouflage is not just an adaptation to the physical environment, but to the perception and mind of the viewer. Billions of photons enter the eye every second, so vision reduces the information to only that which is normally useful. Because shortcuts are taken, sensory systems can be manipulated, and this is what camouflage does. Perceived differences in colour and texture are minimised, distinctive features are concealed, false edges are created and the cues the brain uses to group features into recognisable objects are disrupted. Therefore the study of animal camouflage gives us insights to how other species see the world, as well as an explanation for much of the diversity of animal colour and form that we see around us.

Studying animal camouflage therefore has to be fundamentally interdisciplinary, bringing together concepts and tools from evolutionary and developmental biology, perceptual psychology and computer vision. This multi-pronged attack has, in the last decade, transformed our understanding, showing experimentally that specific mechanisms of concealment and disguise, many first postulated in the late 19th century, can work against animals with different visual systems from our own. But this evidence comes from either observational studies or experiments with artificial prey that (appropriately, given the aims) isolate specific mechanisms. What we can't do is point to a real animal and explain its colour pattern.

Consider these issues. Cat species with spotty coats tend to live in forests and/or rest in trees; this suggests that spots are good camouflage in dappled lighting. Humans, at least, do indeed find them hard to detect. Are the different spot patterns of leopard, ocelot and jaguar equally good solutions for such habitats, differing only because of chance effects during their separate evolutionary histories? Or do the different spot patterns represent solutions to subtle differences in the habitats they occupy? Furthermore, are these patterns the optimal camouflage for animals of these sizes in these habitats, or are there other constraints or trade-offs at play?

The reason we have not answered such questions (very general ones in evolutionary biology, about what constitutes evolutionary design or historical constraint) are threefold. First, we have not, until recently, had adequate ways of describing the patterns on animals, or backgrounds, as they would be represented in the brains of other species. Second, such modelling as has been attempted has only been applied to two-dimensional patterns (i.e. 'flat' animals or flat samples of a pattern). Third, we do not have equivalent data for the backgrounds against which animals might be seen (and at all relevant viewing angles). This grant proposal rectifies these shortfalls by applying novel computational methods ('deep learning' of the sort used by Google and their like) across three sub-projects, with different challenges and with different applications. All tackle long-standing, but unanswered, questions about the adaptive value of colour. Furthermore, the results will have direct application in the human domain.

We have chosen three specific experimental systems - snails, cats and humans - because all have a solid background of research on which to build, but the colours operate at different spatial scales, against different viewers and, importantly, with different mechanisms for generating the patterns. All present new opportunities for a new, integrated approach to studying coloration and the interaction between pattern development and evolutionary function.

Our research will also deliver a computational toolkit, and method, that can determine the best (or worst) camouflage for any object in any environment for any viewer (different species or, indeed, machine vision). This should prove useful not only for concealment, but the study of conspicuousness, in biology, advertising, warning signage, protective clothing and other applications.

We pose three questions about camouflage in quite different systems. The reasons these have not been answered are the same and the solution we provide - a computational toolkit and its conceptual basis - has broad applicability within behavioural and sensory ecology, in perceptual psychology and computer vision, and has direct application in the human domain.

The three systems - the colour morphs of the snail Cepaea nemoralis, coat colour in the cat family, and military uniforms - are commonly accepted, in different contexts, as persuasive examples of camouflage. However, the degree to which the patterns of different morphs/species/nations provide camouflage in different habitats, and whether they represent optima for the size of object and constraints of the pattern-generation mechanism, have not been evaluated.

The reason they have not (related to very general questions in biology and engineering about what constitutes design or constraint) are threefold. First, we have not, until recently, had adequate models for describing the patterns on objects, or backgrounds, as they would be encoded in the brains of other species. Second, such modelling as has been attempted has been applied to 2D patterns viewed from one direction. Third, we do not have equivalent data for the backgrounds against which these objects are seen. Our application rectifies these shortfalls by applying the same recipe: acquisition of large-scale datasets of calibrated photos of the objects and their backgrounds, mapping to relevant perceptual spaces, application of state-of-the-art machine learning techniques to determine optima (and degree of departure from optima of real patterns), supplemented by relevant lab and field experiments, and phylogenetic analysis.

The output will be more than answers to questions about camouflage in three high-profile systems; it will be a method of determining the best camouflage, or conspicuous signal, for any object in any environment for any viewer.

We have identified three groups of potential direct impact beneficiaries, and two groups of indirect beneficiaries. Note that a by-product of our research goal of finding the optimal camouflage for different environments is the ability to predict what colour patterns will be most visible. Both concealment and conspicuousness have high relevance for multiple possible users.

Beneficiaries:

1. The military
The most obvious application of camouflage is concealment in military contexts and, with most current warfare being against low-tech adversaries, conventional 'visual camouflage' is still a priority. We have an existing and ongoing relationship with the defence company QinetiQ and the MoD's Defence Science Technology Laboratory, DSTL.

2. The police
The police have interests in not only maximizing visibility in public control and emergency situations, but also reducing detection for specialist units.

3. Road users
Maximising conspicuousness is relevant to cyclists and other road users (e.g. emergency vehicles) as well as signage. We will run workshops (see below) to which we will invite representatives from relevant organisations (emergency services, local councils, the Highways Agency, the Cyclists' Touring Club etc.); this will allow us to inform, but also draw on an extensive (but largely anecdotal) knowledge base concerning conspicuousness of moving and static road vehicles.

4. The general public
Animal coloration is an excellent platform for explaining scientific concepts from evolution to perception. Apart from academic biologists, we would expect to utilise our existing connections with Bristol Zoo, science discovery centres and the BBC Natural History Unit to disseminate our findings. Interested parties will be invited to attend the concealment and/or conspicuity workshops (see below), and we will develop outreach and education material about camouflage. Many people do not connect using camouflage to hide objects/animals with using high-visibility patterns and materials to increase conspicuousness. Enabling visitors to connect these concepts and design patterns that move between the two extremes will be both educational and entertaining.

5. Project researchers
The proposal is highly interdisciplinary. Each post-doc will be encouraged to learn skills, techniques and understanding from the other, and both will grow intellectually, and become more employable, in the process. They will also be encouraged to do the many components of the university's training programme for lecturers that are made available to research pathway staff. Time for writing independent fellowship applications is also built into the plan of work.

Activities:

1. Media and higher education teacher training for the post-docs.

2. We will organise two sets of paired half-day workshops with non-academic interested parties (stakeholders 1 - 3, listed above) in order to exchange information. The first of each paired workshop would be at the beginning of the project in order to establish what information would be useful to the potential beneficiaries and introduce them to our methodology. A second workshop pair would take place once sufficient project data are available: this pair of meetings would be a chance to disseminate our results, and then elicit feedback on our findings from attendees to potentially inform future research directions. One of each pair will focus on concealment (more relevant to the military and defence companies), the other on conspicuousness (more relevant to emergency services / Highways Agency / cyclists etc.).

3. We will also speak at annual 'industry'-specific events (each one associated with one of the sectors), again to disseminate our research.

4. Public engagement: We will present exhibits at science open days and the annual Festival of Nature in Bristol. We will also develop a 'spot the hidden cat' activity for school visits.