Seafloor image compression with large tilesize vector quantization

Autonomous Underwater Vehicles (AUVs) often communicate with scientists on the surface over an unreliable acoustic channel. The challenges of operating in deep waters, over long distances, and with surface ship noise amount to a communication channel with a very low effective bandwidth. This restriction makes transmission of images, even highly compressed images, quite difficult. We present an image compression algorithm designed to convey the gist of an image to surface operators in a very small number of bytes. Our technique divides a large existing database of underwater images into `tiles', and uses these to reconstruct an approximation to new underwater images from a similar domain. We achieve significantly higher compression ratios than conventional image compression techniques, such as JPEG or SPIHT, while still being able to provide useful visual feedback to the surface.