Scalable compression of imagery

Abstract

Recent image compression technologies value functionality as well as compression efficacy. In this dissertation, several compression methodologies that are both efficient and functional are introduced. The first compression system aims to compress three-dimensional data sets. In this system, existing two-dimensional zerotree coding techniques are extended to three dimensions, and are combined with integer wavelet transforms and context-based arithmetic coding. This system creates a unified framework for both lossless and lossy compression of three-dimensional data sets, and generates a scalable codestream. In our experiments, the system provided an average of 23% decrease in file size for representative three-dimensional data sets, compared to the best available two-dimensional lossless compression techniques. Strategies for transmission of JPEG2000 codestreams over packet-erasure channels are introduced next. These strategies are developed by allowing the channel coder to access information on the structure of the codestream generated by the source coder. Experimental results indicate that a carefully designed packetization strategy can provide significant performance improvement, especially when the channel experiences heavy erasures. Finally, utilization of JPEG2000 to compress scientific data is illustrated. In particular, a JPEG2000-based system that can be used to compress electrocardiogram (ECG) data is presented. Although this system utilizes the JPEG2000 image coding algorithm to compress one-dimensional ECG data, experimental results indicate that it compares favorably with specialized ECG compression methods in the literature.