Extreme Compression for Novelty Detection in Bandwidth Constrained Environments

Abstract

Many applications have been launched based on images. More than two hundred thousand images are uploaded to Facebook alone every minute. Accordingly, memory and bandwidth requirements for uploading images keep increasing. In this digital era, image storage and transmission of images over a channel have become more frequent. In addition, many environments exist that have limited memory and low bandwidth capacity. Applications like the Mars Curiosity Rover have extreme constraints on their memory and bandwidth. However, applications of this sort require a great deal of image processing and transmission. The problem is to look for novelty in unmanned places. To detect novelty, we compare all the images with each other that are captured by the machine. Comparison between extremely large images, such as those taken by scientific instruments on remote planets, takes a great deal of processing time, and transmitting those images back to Earth-based facilities is an even greater challenge. To avoid this problem, we propose a solution of extreme compression that forms a signature, solving both storage and transmission problems. These signatures are further used to detect novelty which yields almost same comparison accuracy as before compression. The whole idea reduces the similar data transfers by concentrating on novel data and also on available bandwidth.