A comparison of image statistics of peacock jumping spider colour patterns and natural scenes

The form of arbitrary sexual signals may be driven by the need to be detectable against the background or, alternatively, by selection for efficient processing by the nervous system. This latter alternative is a prediction of the sensory drive hypothesis extended to include efficient coding as a driver of the form of sexual signals. This hypothesis posits that animal visual systems are adapted to process the visual statistics of natural scenes, and that easily-processed stimuli induce a sensation of pleasure in the viewer. In support of this, natural scene statistics have been found to be preferred not only by humans, but by the peacock spider Maratus spicatus. Here we test if male peacock spiders of the highly sexually dimorphic Maratus genus generally (a) evolve colour patterns with image statistics that contrast with the natural background or (b) exploit a potential processing bias by evolving colour patterns with visual statistics similar to those of natural scenes. We analyse and compare multispectral images of male and female spiders of 21 Maratus species and of natural scenes similar to the spiders’ habitat. We find that the image statistics of male patterns diverge from those of natural scenes, whereas the statistics of female patterns do not. Our results support the idea that colour patterns evolve contrasting image statistics to increase conspicuousness and matching image statistics to be camouflaged. Any processing bias for natural scene image statistics in Maratus thus appears to play little role in the evolution of their sexual signals.

Methods
Multispectral imaging

Multispectral imaging was done using a multispectral camera containing three bird-based optical filters (U, M, and L) previously described in Tedore and Nilsson (2019) and two additional filters previously described in Glenszczyk et al. (2021). By taking a weighted sum of these pre-existing camera filters, we calculated computational filters to generate new spectral sensitivities matching the spectral sensitivity of the salticid green receptor, which typically peaks around 530 nm (De Voe 1975; Yamashita and Tateda 1976; Blest et al. 1981; Zurek et al. 2015; Glenszczyk et al. 2021). The computational filter technique is further described in Tedore and Nilsson (2021) and Glenszczyk et al. (2021).