Dwarf cuttlefish, like octopus, are masters of camouflage. To camouflage the cuttlefish brain must solve a hard computational problem; the brain needs to process the visual environment, compute an appropriate pattern, and then orchestrate the expansion and contraction of hundreds of thousands of pigment filled sacs to generate it. How they accomplish this feat is unknown. Furthermore, cuttlefish do not camouflage by simply matching a substrate pixel by pixel; rather they produce an imperfect approximation of the substrate. Despite this they are able to evade detection by predator, prey, and even humans. Here we hypothesize that the cuttlefish brain encodes a low dimensional representation of the visual world from which it generates an appropriate camouflage pattern. It's an open question what visual features the cuttlefish brain encodes. We argue the brain is encoding visual texture. Visual textures are spatially homogenous images with repeated structure and random variation. Visual texture is unique because it can be compactly and efficiently represented by a small set of summary statistics (Portilla \& Simoncelli, 2000). Here we propose these texture statistics as the low dimensional representation used for camouflage. Using a texture synthesis algorithm we show many cuttlefish skin patterns can be well represented by texture statistics. Next, we present a wide array of visual stimuli to freely behaving cuttlefish and show that texture statistics can be used to quantify camouflage and that they capture cuttlefish pattern change better than other large scale representations. This work is a first step in uncovering the neural algorithm underlying cuttlefish camouflage given that the representation the cuttlefish brain uses to generate camouflage patterns constrains the brain's encoding of the visual world and the generation of a skin pattern. Furthermore, understanding how cuttlefish camouflage fools human vision could shed new insights into human visual perception and representation of texture.