Hawazin Ibrahim, summer 2015, "Improvements to Super-Spatial Structure Prediction Algorithm", Computer Science Department, Faculty of Science, University of Western Ontario, Canada.

M.Sc. Thesis Abstract

In the original Super-Spatial Structure Prediction algorithm for lossless image compression for grayscale images, after the prediction stage is done, the prediction residual is encoded with context-based adaptive lossless image coding (CALIC), and use an arithmetic encoder to encode the rest of other information that required to reconstruct the original image back correctly (with no any loss of data), since it is a lossless compression scheme. In fact, these information includes the classification map, the locations of matches, the prediction residual for structure and non-structure blocks, and the signs of predictions residual. Also, in the original work, the authors confirm that the encoder sometimes get broken when it switches between GAP and Super-Spatial Structure Prediction so many times, since the selected predictor is actually based on the type of the current block that been encoding, structure or non-structure. In this work, we focus on reducing the bit rate and compress the image even more. So, we propose a modification for the original Super-Spatial Structure prediction algorithm to further compress the image and fix the encoder's problem that happen with too many switching. Our experimental results show that the compression performance of the proposed improvement scheme outperforms original Super-Spatial Structure prediction algorithm in terms of bit-rate.