Conditional entropy-constrained vector quantization of chromatic information in document images.

Abstract

We present in this thesis original research on Conditional Entropy-Constrained Vector Quantization (CECVQ) of chrominance information in document images. First, we review the existing image compression formats that could be used for document images: JPEG 2000, DjVu, LDF and PNG. Then, we develop the CECVQ algorithm to adapt it to image quantization and we show its excellent distortion-rate performance. Finally, we develop our own coding scheme targeted at the compression of chrominance information and based on the piecewise-constant model. Color space transformation, CECVQ and context-based entropy coding form the skeleton of our scheme. We also show the strengths and the weaknesses of this new coding scheme, and suggest possible improvements and other approaches.