A computational approach to the quantification of animal camouflage

Abstract

Evolutionary pressures have led to some astonishing camouflage strategies in the animal kingdom. Cephalopods like cuttlefish and octopus mastered a rather unique skill: they can rapidly adapt the way their skin looks in color, texture and pattern, blending in with their backgrounds. Showing a general resemblance to a visual background is one of the many camouflage strategies used in nature. For animals like cuttlefish that can dynamically change the way they look, we would like to be able to determine which camouflage strategy a given pattern serves. For example, does an inexact match to a particular background mean the animal has physiological limitations to the patterns it can show, or is it employing a different camouflage strategy (e.g., disrupting its outline)? This thesis uses a computational and data-driven approach to quantify camouflage patterns of cuttlefish in terms of color and pattern. First, we assess the color match of cuttlefish to the features in its natural background in the eyes of its predators. Then, we study overall body patterns to discover relationships and limitations between chromatic components. To facilitate repeatability of our work by others, we also explore ways for unbiased data acquisition using consumer cameras and conventional spectrometers, which are optical imaging instruments most commonly used in studies of animal coloration and camouflage. This thesis makes the following contributions: (1) Proposes a methodology for scene-specific color calibration for the use of RGB cameras for accurate and consistent data acquisition. (2) Introduces an equation relating the numerical aperture and diameter of the optical fiber of a spectrometer to measurement distance and angle, quantifying the degree of spectral contamination. (3) Presents the first study assessing the color match of cuttlefish (S. officinalis) to its background using in situ spectrometry. (4) Develops a computational approach to pattern quantification using techniques from computer vision, image processing, statistics and pattern recognition; and introduces Cuttlefish 72x5, the first database of calibrated raw (linear) images of cuttlefish.