Does Disruptive Camouflage Conceal Edges and Features?

Current Zoology 61 (4): 708–717, 2015

Does disruptive camouflage conceal edges and features?

* Richard J. WEBSTER Department of Biology, Carleton University, Ottawa, K1S 5B6, Canada

Abstract Camouflage is ubiquitous in the natural world and benefits both predators and prey. Amongst the range of concealment strategies, disruptive coloration is thought to visually fragment an animal’s’ outline, thereby reducing its rate of discovery. Here, I propose two non-mutually exclusive hypotheses for how disruptive camouflage functions, and describe the visual mechanisms that might underlie them. (1) The local edge disruption hypothesis states that camouflage is achieved by breaking up edge information. (2) The global feature disruption hypothesis states camouflage is achieved by breaking up the characteristic features of an animal (e.g., overall shape or facial features). Research clearly shows that putatively disruptive edge markings do increase concealment; however, few tests have been undertaken to determine whether this survival advantage is attributable to the distortion of features, so the global feature disruption hypothesis is under studied. In this review the evidence for global feature disruption is evaluated. Further, I address if object recognition processing provides a feasible mechanism for animals’ features to influence concealment. This review concludes that additional studies are needed to test if disruptive camouflage operates through the global feature disruption and proposes future research directions [Current Zoology 61 (4): 708–717, 2015]. Keywords Antipredator, Background matching, Contour, Crypsis, Camouflage, Disruptive coloration, Edge detection, Object recognition

Animal camouflage is the phenomenon by which ani- ground perceived by predators at the time and prey’s mals are concealed in plain sight. Offensive camouflage age, and in the microhabitat where the prey is most allows predators to approach their prey without being vulnerable to visually hunting predators” (Endler, 1984). seen (increasing the likelihood of a successful attack), According the success of background matching is en- whereas defensive camouflage allows prey to avoid be- tirely dependent on the visual appearance of the envi- ing seen by their predators (increasing their survival) ronment in which the animal is located (Darwin, 1859; (Cott, 1940; Ruxton et al., 2004; Stevens and Merilaita, Wallace, 1889; Poulton, 1890). Alternatively, disruptive 2011). In addition, there are camouflage strategies that coloration consists of markings that breakup and distort provide concealment for auditory, chemical, tactile, and edge information; i.e. “… a set of markings that creates visual sensory modalities (Ruxton, 2009). Not surpris- the appearance of false edges and boundaries and ingly, there is typically a selective advantage to being hinders the detection or recognition of an object’s, or well camouflaged. A classical evolutionary example is part of an object’s, true outline and shape” (Stevens and the melanistic moths change in colour frequency, with Merilaita, 2009). Although background matching and the dark morphs becoming more common as they had disruptive coloration are thought to be distinct camouf- better camouflage on trees that had become darkened by lage strategies ((Merilaita, 1998; Cuthill et al., 2005) pollution (Kettlewell, 1956; Kettlewell and Conn, 1977). and references therein), an animal can improve its ca- This review focuses on visual camouflage, in particular mouflage by employing a combination of the two. In- the phenomenon of disruptive coloration, its relation- deed, disruptive coloration requires some degree of ship to background matching, and how the visual pro- background matching to be effective (Cott, 1940; Fraser, cessing of disruptive camouflage makes it a unique ca- et al., 2007; Troscianko et al., 2013; Kang et al., 2014). mouflage strategy. Since both offensive and defensive camouflage have Both camouflage strategies, background matching and evolved as protective mechanisms to avoid visual per- disruptive coloration, are thought to be ubiquitous in ception by receivers, it is essential to consider camouf- nature. Background matching occurs when an animal’s lage from the receiver’s perspective (e.g. a sensory eco- colour pattern (a composite of its luminance, coloration, logy approach) (Bennett and Cuthill, 1994; Endler and and texture) resembles “a random sample of the back- Basolo, 1998; Stevens and Merilaita, 2011).

Webster RJ: Disrupting edges and features 709

An important question in camouflage research is to The local edge disruption hypothesis proposes that determine what type of visual information is exploited concealment is achieved by breaking up of edges, a by disruptive coloration (Troscianko et al., 2009). Does concept that has strong empirical support (Stevens and it operate by concealing the entire animal (i.e. by dis- Cuthill, 2006). Markings that breakup an animal’s edges rupting the animal’s entire outline), or, instead, by con- make the animal less visible because their edges are cealing discrete portions of its edge that form highly discontinuous, which degrades the boundary between recognizable structures (i.e. ‘features’ such as an eye or the animal and its environment. Stevens and Cuthill (2006) ear)? This is a vital question because each scenario and Webster et al. (2013) used artificial moth targets to likely interferes with different levels of perceptual or- show that edge markings make targets harder to detect ganization (Wagemans, 2015). Researchers are currently (as quantified by computer vision algorithms using a uncertain if masking feature information contributes to vision model of edge detection). Kang et al. (2015) pro- camouflage, or if this type of information takes too long vided further insight into the importance of edge visibi- to be processed to have any meaningful influence on lity for animal concealment in the field. They showed camouflage tasks. that moth re-positioning behaviour increases the break- To begin this discussion we require clear definitions up of their edges, inter alia, thereby enhancing their of edges and features. An ‘edge’ is a line or curve that is camouflage. This experiment also demonstrated that a the boundary between where the animal ends and the small increase in the amount of boundary that is broken background begins and is visible due to a sharp discon- up produces a substantial increase in concealment. tinuity in appearance between the two surfaces (Elder The second hypothesis is the global feature disrup- and Velisavljevic, 2009, and references therein). Such tion hypothesis which proposes that disruptive colora- edge detection relies on spatial contrast between animal tion produces concealment by obscuring the outlines of and background and occurs early in perceptual organi- characteristic features of an animal. For global feature zation (Perrinet and Bednar, 2015). In contrast, a feature disruption, markings are distributed to intersect distinc- is defined as object-related information that is used for tive features (e.g. the corners of triangle-like moths to recognition. For example, if searching for a cat, then the impair predators shape perception). Alternatively, for collection of edges that resemble a cat’s face will con- local edge disruption, markings would be distributed tribute to the cat’s visibility in the scene. Such features either randomly (assuming all edge fragments equally influence detection) or to break the long edge fragments likely contribute to the search images used by animals (Panis and Wagemans, 2009). Both hypotheses are com- when visually seek familiar objects (sensu (Lawrence pared in Figure 1. Evidentially, these two hypotheses and Allen, 1983)). While much is still unknown about are not mutually exclusive; rather, they are suggestive how object recognition contributes to visual search and of the spectrum of possible sources of visual informa- the possible role of gestalt principles (Wagemans et al., tion (Panis and Wagemans, 2009) that disruptive camou- 2012a; Wagemans et al., 2012b), what is clear is that flage might exploit. beyond edge detection, grouping of edge elements is Since edge and feature disruption are not mutually required. The visual process responsible for this aggre- exclusive (and it is challenging to identify the source of gating of edge elements into shapes is called ‘contour visual information that is being exploited by the re- integration and completion’ (and is described more com- ceiver), there have been few empirical studies that deal pletely later). These definitions suggest that there might specifically with feature disruption. Compelling evi- be a difference between camouflaging edges and fea- dence for the role of feature disruption is that having tures. Further it could be possible that different types of additional edge markings has been show to, not only disruption exist in different ecological contexts. improve concealment, but to make the target harder to 1 Disruptive Coloration: Edges versus recognise (Webster et al., 2013). If edge markings only Features provide concealment by reducing edge detection, then ability to recognize these targets would not change in The differences between features and edges have led the presence/absence of edge markings (i.e., edge mar- to the proposal of two hypotheses for explaining how kings would only increase detection time). Alternatively, disruptive coloration might function: the local edge the global feature disruption hypothesis predicts that as disruption hypothesis and the global feature disruption features are increasingly broken up, it will become more hypothesis.

710 Current Zoology Vol. 61 No. 4

Fig. 1 This diagram illustrates how different portions of animals’ boundary convey feature information The first column represents all the edges possibly detected. The second column represents a randomly broken up animal boundary (e.g. by edge markings). Finally the third column breaks up key portions of the animals’ boundary to prevent feature detection. Feature disruption prevents top-down information contributing to the visibility of an animal, by breaking up corner points and facial features. difficult to recognise the target. Webster et al. (2013) misclassification have a sizeable impact on survivorship. demonstrated this using eye-tracking. Targets with bro- Importantly, there is a speed-accuracy trade-off while ken up outlines were more likely to be misclassified foraging, with there being a high cost associated with (seen-and-overlooked), and took longer to be discov- resampling an area during visual search. Therefore, a ered during final inspection. This suggests that a moth’s small increase in the being seen-and-overlooked is ex- outline acts as a ‘feature’, since its visibility influences pected to lead to large gains in survivorship, especially recognition, which is an important criteria for the global in an ecological context where the hunter but must fora- feature disruption hypothesis. ge throughout a large landscape (as opposed to being As Webster et al. (2013) is unique in testing the role limited to a small computer screen). Taken together, the of recognition processing /misclassification of disrup- work done by Webster et al. (2013) offers compelling tive targets, it is worth considering this works possible evidence that misclassification is increased by edge limitations. One critique of the study is if targets are disruptions and aids concealment. truly misclassified when the predator foveates near the To date mixed results have been reported regarding target; it is possible that the predator might not be in- whether or not the global disruption hypothesis is sup- specting the moth target, but instead inspecting an adja- ported. I predicted that the use of edge versus feature cent patch of the background. If so, the misclassification disruption depends on the target itself; for example, a metrics used by Webster et al. (2013) would include triangular moth with straight edges has a very visible additional noise. While feasible, this scenario is unlikely outline and would greatly improvement its concealment in this experiment because the targets were a relatively with edge disruption camouflage; whereas, a triangular large in the predators’ field of view (3.9° × 2.2°). Ano- moth with ruffled edges already has an outline that is ther critique is that there is only a slight increase in less visible and would therefore benefit most from dis- proportion of targets seen-and-overlooked. The question ruptive camouflage of its features. Indeed, in order to that this raises is: does this change have a meaningful produce evidence for the global feature disruption hy- survivorship effects? I argue that even small changes in pothesis, it is necessary to show that a camouflage pat-

Webster RJ: Disrupting edges and features 711 tern’s function is conditional upon the visibility of a fea- false boundary that is more visible than the animal’s ture. Along these lines, a recent study conducted by our limbs (Cuthill and Szekely, 2009); and disruption of laboratory varied both the presence or absence of edge surface that breaks up the uniformity of an object, ma- markings, and the target’s (1) shape complexity or (2) king it appear disjointed (Stevens et al., 2009; Sey- boundary visibility (Webster, et al., 2015). The results moure and Aiello, 2015). These types of disruption are of this experiment demonstrated that the ability of edge described by Stevens and Merilaita (2009), but others markings to provide concealment depends on the visi- may be added to the list, like face disruption (Cott, bility of the targets’ boundary, but does not depend on 1940). If certain types of disruptive coloration empha- the targets’ shape. This suggests that the local edge dis- size masking of feature visibility (e.g. coincidental disruption hypothesis may be favored over the global fea- ruption and face disruption), then they might be ex- ture disruption hypothesis in this paradigm. While this pected to occur in ecological contexts where masking approach provides insight into which source of visual features is more advantageous. information is driving camouflage, care must be taken in interpreting results from artificial animal targets. A 3 Visual Mechanisms of Disruptive critique of this and other similar studies is that the Coloration shapes used for the artificial animal target lack ecologi- The local edge disruption hypothesis proposes a cal realism. In order to show that feature concealment is similar mechanism to that outlined in: (Troscianko et al., an important component of disruptive coloration the use 2009); therefore, the focus here will be the possible of more realistic features will be necessary. For instance, mechanisms for the proposed global feature disruption the boundaries and shapes of targets should be based on hypothesis. the local and global statistics of animal outlines (Fründ and Elder, 2013). A common misconception of visual search is the or- Although edge and feature disruption likely operate der in which different sources of visual information are together, there are some ecological contexts in which processed. This misconception can be addressed by feature concealment might be favored over edge disrup- considering the following: if feature detection is rapid, tion and vice versa. For instance, when a receiver has early-acting, and contributes to visual search, then con- the visual task of spotting a diverse array of prey items cealment of features is a feasible camouflage strategy. with inconsistent features, then the prey item would Alternatively, if feature detection is slow, late-acting, most benefit from reducing edge distinctiveness. Alter- and does not contribute to visual search, then we should natively, for a receiver that has the visual task of spot- have less confidence in this hypothesis. ting one type of prey item, the prey item would most Two commonly used terms in vision science and benefit from concealing those features that are vulnera- psychology are ‘bottom-up’ and ‘top-down’. Bottom-up ble to being recognized. Similarly, a predator that is the processing uses mid-level vision and surface informa- dominant risk to its prey species would benefit from tion to order local geometric parts and depth cues having a camouflage strategy that prevents its prey from (Wilson and Keil, 2001). Such bottom-up cues for de- detecting its features. Accordingly, some ecological tecting camouflaged animal include searching for tex- contexts may be better suited to either exploiting edges tures and colour that make the animal different from its or features. background. Conversely, top-down analysis emphasizes object-related information that is important to the task at 2 Types of Disruptive Markings hand (Wilson and Keil, 1999). Such top-down cues for There are many possible types of disruptive markings detecting camouflaged animal include searching for including: disruptive marginal patterns that interfere characteristic shape of the animal outline or face. Top- with boundary detection to mask outlines (Merilaita, 1998; down information rapidly contributes to discovering Cuthill et al., 2005; Merilaita and Lind, 2005; Cuthill, et objects and in some cases: “as soon as you know it is al., 2006; Schaefer and Stobbe, 2006; Stevens and Cu- there, you know what it is” (Grill-Spector and Kanwish- thill, 2006; Stevens et al., 2006; Fraser et al., 2007; Di- er, 2005). Importantly, top-down cues affect how noti- mitrova and Merilaita, 2009; Stevens et al., 2009; Tros- ceable an object is, therefore, masking feature informa- cianko et al., 2013; Webster et al., 2013); coincidental tion allows for increased concealment. Importantly, disruption, where high-contrast markings on adjacent top-down and bottom-up information are processed si- body limbs align through behavioural posture to create a multaneously (not sequentially) (Kotowicz et al., 2010).

712 Current Zoology Vol. 61 No. 4

Locating targets is faster when you prioritize features first step of object recognition, recent developments relevant to the task (e.g. key-likeness for recovering lost from vision science may explain how local edge disrup- keys, or a prey’s characteristic colour and shape when tion is able to conceal an animal. It is known that the hunting). It has been shown that, using viewing scenes greater the amount of an animal’s boundary that is visi- with peripheral vision cues alone, an animal’s presence ble, the more conspicuous the animal is. An animal’s can be determined within 150 milliseconds by a human boundary is assembled by a receiver’s visual system ‘predator’ (Thorpe et al., 2001; VanRullen and Thorpe, using visible edge fragments (Osorio and Srinivasan, 2001; Kirchner and Thorpe, 2006). As an object’s fea- 1991). This assembly of edge fragments into predictable tures rapidly influence that object’s visibility, it is likely arrangements is fundamental to how visual systems that disrupting an animal’s features will play an impor- processes visual information (Geisler et al., 2001). Inte- tant role in camouflage, especially during certain tasks. restingly, there is evidence to suggest that simple in- Critically, these differences in bottom-up and top-down formation, such as edge fragments, plays a role in object processing provide a framework to discuss what sources recognition. Because the edges that make up an ani- of visual information disruption exploits. mal’s outline are more curved than the edges of non- Not only is top-down information effective for de- animal stimuli, it is possible to detect and recognize an tecting animals, but camouflaged animals are at a higher animal from the shape of its edge alone (Perrinet and risk of being discovered when receivers place more em- Bednar, 2015). The advantage of using edge informa- phasis on top-down information. As visual tasks change, tion to find an animal in a scene is that it is computa- so too does the way by which top-down and bottom-up tionally straightforward (Bergevin and Levine, 1992). information is prioritized. When free viewing (visually Further, this type of information is processed early by exploring a scene with no task), bottom-up information mammalian visual systems (and probably those of other is processed earlier; alternatively, when performing an taxa), therefore allowing for rapid decision making active search (searching for a known target during vi- based on assembly of edge fragments (Perrinet and gilance bouts), features similar to the target (top-down Bednar, 2015). information) are processed earlier (Williams, 1966; Zo- The limitation of camouflaging edges is that edge in- hary and Hochstein, 1989; Wolfe, 1994; Motter and formation only distinguishes an animal from its back- Belky, 1998; Beutter et al., 2003; Najemnik and Geisler, ground, but does not distinguish it from other animals. 2005; Rutishauser and Koch, 2007). There is likely high Indeed, it is unlikely that edge information alone can selection for animals to mask their features, especially distinguish predatory animals from other less relevant as these features are more likely to attract attention and animal stimuli. Therefore, local edge information ig- affect decision making during vigilance and predatory nores context. Often there is a need to recognize speci- foraging behaviours. fic species that poses a high selective pressure (i.e. it is worth recognizing features of a key predator, if that will 4 Object Recognition Processing and allow for early detection and avoidance). In such cases, Disruptive Camouflage recognizing familiar features is advantageous. While edge information can be valuable for detecting animals, In order to describe how top-down information is if recognizing particular features of a dominant predator incorporated into decision making, it is first necessary provides a further advantage, then camouflage resear- to briefly introduce the fundamentals of object recogni- chers need to give more consideration to how visual tion processing. The processes for object recognition mechanisms like perceptual grouping influence feature can be coarsely divided into three stages: (i) edge detec- visibility and contribute to disruption. tion, (ii) feature grouping, and (iii) object identification/ Perceptual grouping: The second stage of object reco- classification (Palmer, 1999). While much is unknown gnition is contour integration. Here, edge information is about object recognition processes between taxa, the assimilated into shapes by the mechanism of perceptual above steps are thought to be common across vertebra- grouping (Fig. 2A); importantly, the feature detection tes irrespective of anatomical differences (Nielsen and hypothesis depends on perceptual grouping (for extra- Rainer, 2007). Below, each of these stages of recogni- polation of edge information into features). Perceptual tion is considered in the context of disruptive camouf- grouping is the task of assigning visual elements to in- lage. dividual objects, a phenomenon that is well documented Edge detection: Since edge detection is an important in humans (Wertheimer, 1923; Helson, 1933; Wagemans

Webster RJ: Disrupting edges and features 713 et al., 2012a; Wagemans et al., 2012b). Perceptual circle fragments are rotated (to prevent perceptual grou- grouping can also produce visual illusions (Halko et al., ping of elements into a circle) they employed a very dif- 2008), which has been documented across animal ferent camouflage pattern (Zylinski et al., 2012). Taken groups: mammals (Vonderheydt et al., 1984; Fagot and together, it appears that perceptual grouping is likely to Tomonaga, 2001; Spinozzi et al., 2009), birds (Zanforlin, have an important influence on camouflage. 1981; Nieder and Wagner, 1999), fish (Wyzisk and Perceptual grouping leads to outline integration, which Neumeyer, 2007; Sovrano and Bisazza, 2008), and in- is not always a simple task because there are often por- sects (Horridge et al., 1992)). If visual illusions, in- tions of an object’s boundaries that are hidden from volving the perception of boundaries, are a byproduct of view, thereby forcing the observer to fill-in-the-blanks perceptual grouping, then the commonness of these (Elder et al., 2003). Incomplete outlines are particularly visual illusions across animal taxa suggests that percep- common in natural scenes with camouflaged animals. tual grouping is also widespread. An example of how An animal’s outline can be hidden from view by: insuf- perceptual grouping affects camouflage has been docu- ficient animal-to-background contrast (Marr and Hil- mented in cuttlefish (Sepia officinalis; Fig. 2B). Cuttle- dreth, 1980; Stevens and Cuthill, 2006; Kang, et al., fish are renowned for rapid colour changing to suit their 2014), or by partial occlusion of the animal by fore- background. Zylinski et al. (2012) found that on background objects (Tvardikova and Fuchs, 2010). To deal grounds of full and fragmented circles the cuttlefish with incomplete boundary information, object recogni- choose one camouflage pattern; however, when the same tion has been shown to ‘fill-in’ the missing pieces to estimate objects’ boundaries (a process called contour completion; Fig. 3) (Kellman and Shipley, 1991; An- derson et al., 2002; Anderson, 2013). Since this is a ubiquitous problem, it is not surprising that contour completion is a common operation for feature detection that has been reported in: mammals (Kanizsa et al., 1993; Sato et al., 1997; Deruelle et al., 2000; Fujita, 2001; Parron and Fagot, 2007; Kurylo, 2008; Barbet and Fagot, 2011), birds (Regolin and Vallortigara, 1995; Nagasaka et al., 2005; Cavoto and Cook, 2006; Naka- mura et al., 2010), and fish (Sovrano and Bisazza, 2008; Truppa et al., 2010), but a few experiments suggest some animal do not (Burke et al., 2001). Possibly contour integration and completion mechanisms similarity between humans and birds may account for the surpri- singly consistent results when searching for disruptive targets (Cuthill et al., 2005; Fraser et al., 2007). Object recognition: The final stage of object recognition is to utilize feature information for object identification. Since Thayer (1909), a body of evidence has

Fig. 2 Visual illusions of shape that are perceived by hu- emerged suggesting that shape features are key for ob- mans and non-humans ject recognition (Delvenne and Dent, 2008; Liebe et al., Such fundamental aspects of shape perception by cognitive systems 2009; Soto and Wasserman, 2012), and particularly for are important in the context of disruptive coloration. A. Kanizsa illu- animal recognition by humans (Lloyd-Jones and Luck- sion where shape perception is affected by perceptual grouping, B. hurst, 2002; Elder and Velisavljevic, 2009; Lloyd-Jones Graph adapted from Zylinski et al. (2012) illustrating that cuttlefish dynamically express camouflage strategies in accordance to shape et al., 2010). Further, recognition of shapes can contri- perception of background. There were five background treatments bute to detectability, independent of an object’s colour presented, one plain and four with repeated visual elements (a circle, (Brown et al., 1992; Lehrer and Campan, 2004; Carlile four lines arranged like a circle, four lines arranged randomly [not like et al., 2006; Ings et al., 2012; Schluessel et al., 2012). a circle] and one line). Interestingly, cuttlefish camouflage choice was consistent for backgrounds with circles and broken up circles. This This concept has implications for the assumption that suggests that cuttlefish use perceptual grouping and it affects their disruptive camouflage can be achieved independently of camouflage decision making. background matching (Schaefer and Stobbe, 2006); for

714 Current Zoology Vol. 61 No. 4

Fig. 3 Contour completion: from detected edges to global shape A. Natural viewing condition of a blue jay Cyanocitta cristata, with portions of its edge not detected due to low contrast with background (i.e. its beak) or occlusion from foreground branches (i.e. its underbelly), B. Contour integration (green dots) groups detected edges into a partially com- plete boundary, C. Contour completion (red dots) fills in the missing boundary information (due to low contrast animal-background boundaries and occlusion) to form a closed shape estimation of the animal. instance, a small loss of background colour matching mains an area of active investigation. Future work ought could be tolerated if the change derives a larger benefit to test if the precise placement of disruptive markings from degrading shape recognition. on a boundary determines if edge or feature information As the above examples serve to demonstrate, there is (or both) are concealed. To reduce the visibility of edges, some evidence that the visual systems of humans and long straight edges must be fragmented because other- non-humans share object recognition processes (e.g. wise they form conspicuous boundaries against the perceptual grouping) that could underlie the global fea- background. Alternatively, to delay feature detection, ture disruption hypothesis. Indeed, while mammals and specific parts of an edge should be disrupted. For in- birds have very different brain anatomies (Hodos, 1993), stance, edge markings that intersect specific anatomical feature detection, i.e. recognizing an object’s shape, is features work to degrade shape information; e.g. pig- present in both taxa (Logothetis et al., 1995; Peissig, et mentation placed on the corners of a triangular moth, al., 2005) and operates in a similar manner in each (Nie- eye stripes that disrupt the highly recognizable circular lsen and Rainer, 2007). This similarity of edge/feature outline of an eye, and coincidental disruption that brea- perception across taxa suggests that disruptive colora- ks up an animal’s limbs and body. The field is awaiting tion is a widespread camouflage strategy. an experimental test that can resolve an object’s edge It seems feasible that feature information could con- distinctiveness (relative to edges detected in the back- tribute to visual search performance, and be an influen- ground) from the features that contribute to its recogni- tial factor for camouflage. Familiar objects can be ra- tion. Only by testing how the disruption of precise por- pidly detected by recognizing key features using top- tions of animals’ boundaries alters concealment (Fig. 1) down information. Precisely how disruption affects fea- will we be able to understand which sources of informa- ture detection is unclear and the particular mechanism tion disruptive camouflage exploits. will likely vary across taxa and between visual environ- Another requirement for future studies is the use of ments. This said, disruptive camouflage is likely to ex- more realistic animal targets. To date, the majority of ploit a broad range of processes that are involved in perce- the animal-like targets used to study disruptive colora- ptual grouping (Espinosa and Cuthill, 2014) and contour tion lack naturalistic detail. While it is not required that completion, which are vital stages in object recognition. the animal-like target is a perfect representation of a 5 Future Work single species, it is important that animal targets used to test ideas about disruption have local edge and global The field of camouflage by disruptive coloration re- shape statistics that are realistic since they play an im-

Webster RJ: Disrupting edges and features 715 portant role in directing visual search (Fründ and Elder, tures for object recognition. Pattern Recognition 25: 319–334. 2013, Perrinet and Bednar, 2015). Animals are under Beutter BR, Eckstein MP, Stone LS, 2003. Saccadic and perceptual performance in visual search tasks. I. Contrast detection strong selection for visual systems that detect animals’ and discrimination. Journal of the Optical Society of America - edge fragment and features. Without these realistic sour- Optics Image Science and Vision 20: 1341–1355. ces of information it may be difficult to assess the dual Brown MM, Kreiter NA, Maple JT, Sinnott JM, 1992. Silhouettes role of edge and feature camouflage. As our community elicit alarm calls from captive vervet monkeys Cercopithecus transitions toward using more compelling animal targets, aethiops. Journal of Comparative Psychology 106: 350–359. it is likely that a greater survival benefit of disruptive Burke D, Everingham P, Rogers T, Hinton M, Hall-Aspland S, 2001. Perceptual grouping in two visually reliant species: Hu- camouflage will be revealed. mans Homo sapiens and Australian sea lions Neophoca cine- 6 Conclusion rea. Perception 30: 1093–1106. Carlile PA, Peters RA, Evans CS, 2006. Detection of a looming There is currently a limited amount of experimental stimulus by the Jacky dragon: Selective sensitivity to characte- evidence to explain how (and indeed if) disruptive co- ristics of an aerial predator. Animal Behaviour 72: 553–562. Cavoto BR, Cook RG, 2006. The contribution of monocular depth loration prevents feature detection. One study has shown cues to scene perception by pigeons. Psychological Science 17: that disruptive targets take longer to recognize and are 628–634. misclassified more often (Webster et al., 2013), yet the Cott HB, 1940. Adaptive Coloration in Animals. London: Me- question remains: does disruptive camouflage impair thuen. visibility by exploiting local edge information, global Cuthill IC, Stevens M, Sheppard J, Maddocks T, Parraga CA et al., 2005. Disruptive coloration and background pattern matching. feature information, or both? This review has explored Nature 434: 72–74. some of the possible mechanisms for how disruptive Cuthill IC, Stevens M, Windsor AMM, Walker HJ, 2006. The camouflage might work to conceal edges and features. effects of pattern symmetry on detection of disruptive and Since visual mechanisms rapidly allow top-down in- background-matching coloration. Behavioral Ecology 17: formation to contribute to visual search, it seems con- 828–832. Cuthill IC, Szekely A, 2009. Coincident disruptive coloration. ceivable that disruptive camouflage works by prevent- Philosophical Transactions of the Royal Society B-Biological ing feature detection. Additional mechanistic studies of Sciences 364: 489–496. disruptive coloration should reveal how disruption is Darwin CR, 1859. On The Origin of Species by the Means of distinct from background matching and will elucidate Natural Selection. London: John Murray. the wider implications of how camouflage contributes Delvenne JF, Dent K, 2008. Distinctive shapes benefit short-term memory for color associations, but not for color. Perception & to antipredator defense. Psychophysics 70: 1024–1031. Deruelle C, Barbet I, Depy D, Fagot J, 2000. Perception of partly Acknowledgments The author greatly acknowledges Tom occluded figures by baboons Papio papio. Perception 29: Sherratt, Jean-Guy Godin, Sophie Potter, Changku Kang, Ca- 1483–1497. sey Peet-Paré, Tammy Duong, and Francina Webster for help- Dimitrova M, Merilaita S, 2009. Prey concealment: Visual back- ful comments on the manuscript. During review process this ground complexity and prey contrast distribution. Behavioral paper was further improved thanks to Will Alan, an anonym- Ecology 21: 176–181. ous reviewer, and handling editor Martin Stevens. Several Elder JH, Krupnik A, Johnston LA, 2003. Contour grouping with concepts developed in this paper were influenced by RJW prior models. IEEE Transactions on Pattern Analysis and Ma- attendance of the Gestalt Revision Summer School, organized chine Intelligence 25: 661–674. by Johan Wagemans. Elder JH, Velisavljevic L, 2009. Cue dynamics underlying rapid detection of animals in natural scenes. Journal of Vision 9: 1– References 20. Endler JA, 1984. Progressive background in moths, and a quantita- A nderson BL, 2013. Filling-in the gaps in models of completion. tive measure of crypsis. Biological Journal of the Linnean So- Cognitive Neuroscience 4: 48–49. ciety 22: 187–231. Anderson BL, Singh M, Fleming RW, 2002. The interpolation of Endler JA, Basolo AL, 1998. Sensory ecology, receiver biases and object and surface structure. Cognitive Psychology 44: 148– sexual selection. Trends in Ecology & Evolution 13: 415–420. 190. Espinosa I, Cuthill IC, 2014. Disruptive Colouration and percep- Barbet I, Fagot J, 2011. Processing of contour closure by baboons tual grouping. PLoS ONE 9: e87153. Papio papio. Journal of Experimental Psychology-Animal Fagot J, Tomonaga M, 2001. Effects of element separation on Behavior Processes. 37: 407–419. perceptual grouping by humans Homo sapiens and chimpan- Bennett ATD, Cuthill IC, 1994. Ultraviolet vision in birds: What zees Pan troglodytes: Perception of Kanizsa illusory figures. is its function? Vision Research 34: 1471–1478. Animal Cognition 4: 171–177. Bergevin R, Levine MD, 1992. Extraction of line drawing fea- Fraser S, Callahan A, Klassen D, Sherratt TN, 2007. Empirical

716 Current Zoology Vol. 61 No. 4

tests of the role of disruptive coloration in reducing detectabili- 125. ty. Proceedings of the Royal Society B-Biological Sciences Lloyd-Jones TJ, Luckhurst L, 2002. Outline shape is a mediator 274: 1325–1331. of object recognition that is particularly important for living Fründ I, Elder J, 2013. Statistical coding of natural closed con- things. Memory & Cognition 30: 489–498. tours. Journal of Vision 13: 119. Logothetis NK, Pauls J, Poggio T, 1995. Shape representation in Fujita K, 2001. Perceptual completion in rhesus monkeys Macaca the inferior temporal cortex of monkeys. Current Biology 5: mulatta and pigeons Columba livia. Perception & Psycho- 552–563. physics 63: 115–125. Marr D, Hildreth E, 1980. Theory of edge detection. Proceedings Grill-Spector K, Kanwisher N, 2005. Visual recognition: As soon of the Royal Society of London B 207: 187–217. as you know it is there, you know what it is. Psychological Merilaita S, 1998. Crypsis through disruptive coloration in an Science 16: 152–160. isopod. Proceedings of the Royal Society of London Series B- Halko MA, Mingolla E, Somers DC, 2008. Multiple mechanisms Biological Sciences 265: 1059–1064. of illusory contour perception. Journal of Vision 8: 1–17. Merilaita S, Lind J, 2005. Background-matching and disruptive Helson H, 1933. The fundamental propositions of Gestalt psy- coloration, and the evolution of cryptic coloration. Proceed- chology. Psychological Review 40: 13–32. ings of the Royal Society B-Biological Sciences 272: 665– Hodos W, 1993. The visual capabilities of birds. In: Zeigler HP, 670. Bischof HJ eds. Vision, Brain and Behaviour in Birds. Boston: Motter BC, Belky EJ, 1998. The guidance of eye movements MIT Press, 183–191. during active visual search. Vision Research 38: 1805–1815. Horridge GA, Zhang SW, Ocarroll D, 1992. Insect perception of Nagasaka Y, Hori K, Osada Y, 2005. Perceptual grouping in pig- illusory contours. Philosophical Transactions of the Royal So- eons. Perception 34: 625–632. ciety of London Series B-Biological Sciences 337: 59–64. Najemnik J, Geisler WS, 2005. Optimal eye movement strategies Ings TC, Wang MY, Chittka L, 2012. Colour-independent shape in visual search. Nature 434: 387–391. recognition of cryptic predators by bumblebees. Behavioral Nakamura N, Watanabe S, Betsuyaku T, Fujita K, 2010. Do ban- Ecology and Sociobiology 66: 487–496. tams Gallus gallus domesticus experience amodal completion? Kang C, Stevens M, Moon J-Y, Lee S-I, Jablonski PG, 2014. Ca- An analysis of visual search performance. Journal of Compara- mouflage through behavior in moths: The role of background tive Psychology 124: 331–335. matching and disruptive coloration. Behavioral Ecology 26: Nieder A, Wagner H, 1999. Perception and neuronal coding of 45–54. subjective contours in the owl. Nature Neuroscience 2: 660– Kanizsa G, Renzi P, Conte S, Compostela C, Guerani L, 1993. 663. Amodal completion in mouse vision. Perception 22: 713–721. Nielsen KJ, Rainer G, 2007. Object recognition: similar visual Kellman PJ, Shipley TF, 1991. A theory of visual interpolation in strategies of birds and mammals. Current Biology 17: R174– object perception. Cognitive Psychology 23: 141–221. R176. Kettlewell HBD, 1956. A resume of investigations on the evolu- Osorio D, Srinivasan MV, 1991. Camouflage by edge enhance- tion of melanism in the Lepidoptera. Proceedings of the Royal ment in animal coloration patterns and its implications for Society B-Biological Sciences 145: 297–303. visual mechanisms. Proceedings of the Royal Society of Lon- Kettlewell HBD, Conn DLT, 1977. Further background-choice don Series B-Biological Sciences 244: 81–85. experiments on cryptic lepidoptera. Journal of Zoology 181: Palmer SE, 1999. Vision Science: Photons to Phenomenology. 371–376. Boston: MIT Press. Kirchner H, Thorpe SJ, 2006. Ultra-rapid object detection with Panis S, Wagemans J, 2009. Time-course contingencies in per- saccadic eye movements: Visual processing speed revisited. ceptual organization and identification of fragmented object Vision Research 46: 1762–1776. outlines. Journal of Experimental Psychology -Human Percep- Kotowicz A, Rutishauser U, Koch C, 2010. Time course of target tion and Performance 35: 661–687. recognition in visual search. Frontiers in Human Neuroscience Parron C, Fagot J, 2007. Comparison of grouping abilities in 4: 11. humans Homo sapiens and baboons Papio papio with the Eb- Kurylo DD, 2008. Effects of visual cortex lesions on perceptual binghaus illusion. Journal of Comparative Psychology 121: grouping in rats. Behavioural Brain Research 190: 182–188. 405–411. Lawrence ES, Allen JA, 1983. On the term 'search image'. Oikos Peissig JJ, Young ME, Wasserman EA, Biederman I, 2005. The 40: 313–314. role of edges in object recognition by pigeons. Perception 34: Lehrer M, Campan R, 2004. Shape discrimination by wasps Pa- 1353–1374. ravespula germanica at the food source: Generalization among Perrinet LU, Bednar JA, 2015. Edge co-occurrences can account various types of contrast. Journal of Comparative Physiology for rapid categorization of natural versus animal images. A - Neuroethology Sensory Neural and Behavioral Physiology Scientific Reports 5. 190: 651–663. Poulton EB, 1890. The Colours of Animals. London: Kegan Paul, Liebe S, Fischer E, Logothetis NK, Rainer G, 2009. Color and Trench, Trübner & Co Ltd. shape interactions in the recognition of natural scenes by hu- Regolin L, Vallortigara G, 1995. Perception of partly occluded man and monkey observers. Journal of Vision 9: 1–16. objects by young chicks. Perception & Psychophysics 57: Lloyd-Jones TJ, Gehrke J, Lauder J, 2010. Animal recognition 971–976. and eye movements the contribution of outline contour and Rutishauser U, Koch C, 2007. Probabilistic modeling of eye local feature information. Experimental Psychology 57: 117– movement data during conjunction search via feature-based

Webster RJ: Disrupting edges and features 717

attention. Journal of Vision 7: 1–20. Camouflage and visual perception. Cambridge: Cambridge Ruxton GD, 2009. Non-visual crypsis: A review of the empirical University Press. evidence for camouflage to senses other than vision. Philoso- Truppa V, Sovrano VA, Spinozzi G, Bisazza A, 2010. Processing phical Transactions of the Royal Society B: Biological Sci- of visual hierarchical stimuli by fish Xenotoca eiseni. Beha- ences 364: 549–557. vioural Brain Research 207: 51–60. Ruxton GD, Sherratt TN, Speed MP, 2004. Avoiding attack: The Tvardikova K, Fuchs R, 2010. Tits use amodal completion in evolutionary ecology of crypsis, warning signals and mimicry. predator recognition: A field experiment. Animal Cognition 13: Oxford: Oxford University Press. 609–615. Sato A, Kanazawa S, Fujita K, 1997. Perception of object unity in VanRullen R, Thorpe SJ, 2001. The time course of visual proce- a chimpanzee Pan troglodytes. Japanese Psychological Re- ssing: From early perception to decision-making. Journal of search 39: 191–199. Cognitive Neuroscience 13: 454–461. Schaefer HM, Stobbe N, 2006. Disruptive coloration provides Vonderheydt R, Peterhans E, Baumgartner G, 1984. Illusory con- camouflage independent of background matching. Proceedings tours and cortical neuron responses. Science 224: 1260–1262. of the Royal Society B - Biological Sciences 273: 2427–2432. Wagemans J, 2015. The Oxford Handbook of Perceptual Organi- Schluessel V, Fricke G, Bleckmann H, 2012. Visual discrimination zation. Oxford: Oxford University Press. and object categorization in the cichlid Pseudotropheus sp. Wagemans J, Elder JH, Kubovy M, Palmer SE, Peterson MA et al., Animal Cognition 15: 525–537. 2012a. A century of gestalt psychology in visual perception: I. Seymoure BM, Aiello A, 2015. Keeping the band together: Evi- Perceptual grouping and figure-ground organization. Psycho- dence for false boundary disruptive coloration in a butterfly. logical Bulletin 138: 1172–1217. Journal of Evolutionary Biology, doi: 10.1111/jeb.12681. Wagemans J, Feldman J, Gepshtein S, Kimchi R, Pomerantz JR et Soto FA, Wasserman EA, 2012. Visual object categorization in al., 2012b. A century of gestalt psychology in visual perception: birds and primates: Integrating behavioral, neurobiological, and II. Conceptual and theoretical foundations. Psychological Bul- computational evidence within a "general process" framework. letin 138: 1218–1252. Cognitive Affective & Behavioral Neuroscience 12: 220–240. Wallace AR, 1889. Darwinism: An Exposition of the Theory Of Sovrano VA, Bisazza A, 2008. Recognition of partly occluded Natural Selection with Some of its Applications. London: objects by fish. Animal Cognition 11: 161–166. Macmillian. Spinozzi G, De Lillo C, Truppa V, Castorina G, 2009. The relative Webster RJ, Godin J-GJ, Sherratt TN, 2015. The role of body use of proximity, shape similarity, and orientation as visual shape and edge characteristics on the concealment afforded by perceptual grouping cues in tufted capuchin monkeys Cebus potentially disruptive marking. Animal Behaviour 104: 197– apella and humans Homo sapiens. Journal of Comparative 202. Psychology 123: 56–68. Webster RJ, Hassall C, Herdman CM, Godin J-GJ, Sherratt TN, Stevens M, Cuthill IC, 2006. Disruptive coloration, crypsis and 2013. Disruptive camouflage impairs object recognition. Bi- edge detection in early visual processing. Proceedings of the ology Letters 9: 1–5. Royal Society B-Biological Sciences 273: 2141–2147. Wertheimer M, 1923. Studies concerning the Theory of Shape. Stevens M, Cuthill IC, Windsor AMM, Walker HJ, 2006. Disrup- Psychologische Forschung 4: 301–350. tive contrast in animal camouflage. Proceedings of the Royal Williams LG, 1966. Effect of target specification on objects fix- Society B -Biological Sciences 273: 2433–2438. ated during visual search. Perception & Psychophysics 1: 315– Stevens M, Merilaita S, 2011. Animal camouflage: Function and 318. mechanisms. Cambridge: Cambridge University Press. Wilson RA, Keil FC, 2001. The MIT Encyclopedia of the Cogni- Stevens M, Merilaita S, 2009. Defining disruptive coloration and tive Sciences. Cambridge, Massachusetts: MIT Press. distinguishing its functions. Philosophical Transactions of the Wolfe JM, 1994. Guided search 2.0: A revised model of visu- Royal Society B - Biological Sciences 364: 481–488. al-search. Psychonomic Bulletin & Review 1: 202–238. Stevens M, Winney IS, Cantor A, Graham J, 2009. Outline and Wyzisk K, Neumeyer C, 2007. Perception of illusory surfaces and surface disruption in animal camouflage. Proceedings of the contours in goldfish. Visual Neuroscience 24: 291–298. Royal Society B - Biological Sciences 276: 781–786. Zanforlin M, 1981. Visual perception of complex forms (anomal- Thorpe SJ, Gegenfurtner KR, Fabre-Thorpe M, Bulthoff HH, ous surfaces) in chicks. Italian Journal of Psychology 8: 1–16. 2001. Detection of animals in natural images using far peri- Zohary E, Hochstein S, 1989. How serial is serial processing in pheral vision. European Journal of Neuroscience 14: 869–876. vision. Perception 18: 191–200. Troscianko J, Lown AE, Hughes AE, Stevens M, 2013. Defeating Zylinski S, Darmaillacq AS, Shashar N, 2012. Visual interpolation crypsis: Detection and learning of camouflage strategies. PLoS for contour completion by the European cuttlefish Sepia offi- ONE 8: e73733. cinalis and its use in dynamic camouflage. Proceedings of the Troscianko T, Benton CP, Lovell PG, Tolhurst DJ, Pizlo Z, 2009. Royal Society B - Biological Sciences 279: 2386–2390.