Visual perception and cuttlefish camouflage
Section snippets
Visual textures and statistical inference
To be useful, visual systems must make the most of noisy biological measurements to solve complex problems quickly. More than being merely difficult, many of these problems are ill-posed, in that they have many possible solutions [1,2]. For example, understanding the 3d layout of a visual scene from sparse 2d images—as retinal images are—is difficult because it is usually underconstrained. In practice, however, we and other animals routinely solve such problems. This is possible because, in the
Describing visual textures
Which image statistics do we use to perceive textures? In 1962, Bela Julesz proposed that textures are perceived in terms of their Nth-order statistics, where N is the number of locations in an image (e.g. defined by the vertices of an N-gon) that need to be compared. Images with matching order statistics below a certain N were hypothesized to be indiscriminable [11]. In what became known as the Julesz conjecture, Julesz later hypothesized that humans are unable to discriminate textures
Evidence for texture perception in animals
How widespread is visual texture perception among animals? Human psychophysics rely on the subjective notion of ‘perceptual equivalence’ to quantify texture similarity. Different criteria are required with animals, but behavioral discrimination studies in insects and cephalopods suggest that texture perception is not limited to mammals or even vertebrates.
Visual perception in honeybees, for example, can be readily probed in the laboratory by pairing visual stimuli of a given class with a
How do animals learn to perceive texture?
If modern deep-learning approaches do perform reasonably well in visual-texture matching tasks [29,32••], they nevertheless typically require hundreds of thousands to millions of training examples. Animals, by contrast, must extract texture statistics with very few examples. Humans become sensitive to the statistics of visual textures over early postnatal development. Simple texture discrimination can be used for object segmentation in 14-week-old infants [38]. At 2–3 months, infant brains show
Cuttlefish as unique models for the study of texture perception
The ability of cephalopods to match their substrate has fascinated students of natural history for millenia [52]. Coleoid cephalopods are a group of molluscs whose ancestral lineage split from our own 580–610 million years ago (protostome-deuterostome divergence). Roughly 200 million years ago, these cephalopods internalized their shells, took up a predatory lifestyle, evolved by far the largest brains amongst invertebrates and developed an active camouflage ability [53]. Cephalopods therefore
Outlook
We have highlighted recent and older work on visual textures, the perception of which appears to be a key component of vision in many different animal groups. This is perhaps most clearly demonstrated by camouflage, which in many instances functions as an adaptation to visual texture perception in others. Cephalopods, possessing active camouflage under neural control, reveal aspects of their perception through their choice of skin pattern in different visual environments. This remarkable
Conflict of interest statement
Nothing declared.
References and recommended reading
Papers of particular interest, published within the period of review, have been highlighted as:
• of special interest
•• of outstanding interest
Acknowledgements
We would like to thank Stephan Junek and Amber Longo for images, Marcel Lauterbach for STED imaging, Claudio Polisseni for μCT imaging, Dominic Evans, Theodosia Woo, and Sara Haddad for their suggestions on the manuscript, as well as Matthias Kaschube, Stephanie Palmer, Jonathan Miller, Kenji Doya, the instructors and students of the Cajal Behavior of Neural Systems course 2018, and the members of the Laurent laboratory for many useful discussions.
References (61)
- et al.
Vision as Bayesian inference: analysis by synthesis?
Trends Cogn Sci
(2006) Visual perception of materials and their properties
Vision Res
(2014)- et al.
When is scene identification just texture recognition?
Vision Res
(2004) - et al.
Building the gist of a scene: the role of global image features in recognition
Prog Brain Res
(2006) - et al.
Texture synthesis using hierarchical linear transforms
Signal Process
(1988) - et al.
Development of sensitivity to texture and contour information in the human infant
J Cogn Neurosci
(2005) Contrast negation and texture synthesis differentially disrupt natural texture appearance
Front Psychol
(2012)- et al.
Motor neurons and the sense of place
Neuron
(2011) - et al.
A theory of multineuronal dimensionality, dynamics and measurement
bioRxiv
(2017) High-level vision as statistical inference
The New Cognitive Neurosciences
(1999)
Computational vision and regularization theory
Nature
Natural image statistics and neural representation
Annu Rev Neurosci
Attentional requirements in a “preattentive” feature search task
Nature
Textons, the elements of texture perception, and their interactions
Nature
A parametric texture model based on deep convolutional features closely matches texture appearance for humans
J Vis
A parametric texture model based on joint statistics of complex wavelet coefficients
Int J Comput Vis
Images, statistics, and textures: implications of triple correlation uniqueness for texture statistics and the Julesz conjecture: comment
J Optic Soc Am A
Textures as probes of visual processing
Annu Rev Vis Sci
Visual pattern discrimination
IRE Trans Inf Theory
Depth estimation from image structure
IEEE Trans Pattern Anal Mach Intell
Statistics of natural image categories
Netw Bristol Engl
A summary statistic representation in peripheral vision explains visual search
J Vis
"ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
International Conference on Learning Representations (ICLR)
On seeing stuff: the perception of materials by humans and machines
Proc SPIE
How camouflage works
Philos Trans R Soc B Biol Sci
Cephalopod chromatophores: neurobiology and natural history
Biol Rev Camb Philos Soc
Identifying the structure in cuttlefish visual signals
Philos Trans R Soc Lond B Biol Sci
Cephalopod Behaviour
On perceptual analyzers underlying visual texture discrimination: part I
Biol Cybern
Decorrelation methods of texture feature extraction
IEEE Trans Pattern Anal Mach Intell
Cited by (11)
Neural control of cephalopod camouflage
2023, Current BiologyFunctional organization of visual responses in the octopus optic lobe
2023, Current BiologyMulti-level control of adaptive camouflage by European cuttlefish
2022, Current BiologyCitation Excerpt :Moreover, previous findings that visually distinct stimuli elicit very similar responses,4,12 and conversely, that essentially identical 2D images can produce different responses depending on the presence of physical depth cues,7 also support the body pattern model. A means by which cuttlefish might implement both feature matching and body pattern models of camouflage selection is suggested by recent findings on their motor system,5,6 where analysis of correlations in spontaneous activity among 17,305 cuttlefish chromatophores implies a multi-level motor control hierarchy. The successive levels in the hierarchy corresponded to small groups of chromatophores, elements of pattern components, pattern components, and—perhaps—body patterns.
Cuttlefish color change as an emerging proxy for ecotoxicology
2023, Frontiers in PhysiologyEfficient processing of natural scenes in visual cortex
2022, Frontiers in Cellular Neuroscience
- 1
Present address: Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan.