Empirical analysis of BWT-based lossless image compression

The Burrows-Wheeler Transformation (BWT) is a text transformation algorithm originally designed to improve the coherence in text data. This coherence can be exploited by compression algorithms such as run-length encoding or arithmetic coding. However, there is still a debate on its performance on images. Motivated by a theoretical analysis of the performance of BWT and MTF, we perform a detailed empirical study on the role of MTF in compressing images with the BWT. This research studies the compression performance of BWT on digital images using different predictors and context partitions. The major interest of the research is in finding efficient ways to make BWT suitable for lossless image compression.;This research studied three different approaches to improve the compression of image data by BWT. First, the idea of preprocessing the image data before sending it to the BWT compression scheme is studied by using different mapping and prediction schemes. Second, different variations of MTF were investigated to see which one works best for Image compression with BWT. Third, the concept of context partitioning for BWT output before it is forwarded to the next stage in the compression scheme.;For lossless image compression, this thesis proposes the removal of the MTF stage from the BWT compression pipeline and the usage of context partitioning method. The compression performance is further improved by using MED predictor on the image data along with the 8-bit mapping of the prediction residuals before it is processed by BWT.;This thesis proposes two schemes for BWT-based image coding, namely BLIC and BLICx, the later being based on the context-ordering property of the BWT. Our methods outperformed other text compression algorithms such as PPM, GZIP, direct BWT, and WinZip in compressing images. Final results showed that our methods performed better than the state of the art lossless image compression algorithms, such as JPEG-LS, JPEG2000, CALIC, EDP and PPAM on the natural images.