Cuttlefish Camouflage Quantification via Novel Neural Network Approaches and Hyperspectral Imaging

Rapid adaptive camouflage is a critical defense mechanism for cephalopods. The characterization of cephalopod camouflage has thus far been reserved almost exclusively to qualitative descriptions, and research on camouflage quantification remains in a nascent state. Qualitative characterizations do not capture the full multifactorial nature of camouflage, nor do they provide a comprehensive metric by which the degree and effectiveness of cephalopod camouflage from the perspective of a given predator can be quantitatively measured in a given scene. Here, I propose a "texture distance" metric that integrates lower and higher dimensional visual features to give a pixel-wise read out of the similarity between a cephalopod's texture and its background texture. This metric is based on a previously developed algorithm that utilizes an artificial neural network to perform texture synthesis. Such a quantifying method would allow researchers to gain more insight into the amount of evolutionary pressure camouflage might exert on predator visual systems. The proposed metric is developed, validated, and used alongside other analyses to investigate the search strategies of human subjects in a camouflage detection task, especially with respect to the difference between human subjects who were successful or unsuccessful in the search task. Furthermore, hyperspectral images (HSI) of cephalopods under natural lighting conditions in the wild were used to measure chromatic and luminance discriminability in the visual color space of a subset of real natural fish predators. This HSI analysis suggests sophisticated color-matching across predator types, while also suggesting cephalopods are more detectable via changes in luminance. This result is compatible with the results of previous studies, but opens the door to exciting new research possibilities.