Fast, robust and adaptive lossless image compression

• Autorzy: Starosolski R.
• Języki publikacji: EN

Abstrakty

EN

For applications like image transmission or storage we need fast and adaptive lossless compression algorithms. A speed improvement must not be achieved at the expense of significant compression ratio deterioration or too big memory requirements. The robustness, which may be defined as a performance on the worst case of data, is very important in practical applications. Presented algorithm uses the traditional decorrelations - statistical compression scheme of adaptive image compression. We introduce many modifications to improve speed and robustness of the algorithm. Firstly, we vastly increase the processing speed by altering the traditional statistical compression scheme. Instead of coding each symbol and updating the data model each time a symol is coded, we update the model only after coding some symbols. We construct a robust family of codes based on the GOLOMB codes and adapted to the real image data - that is to the finite alphabet of not ideally xeponential symbol distribution. In order to quickly adapt to the specific image data model uses a variable number of context buckets and is updated with a variable frequency - starting with a single collective context bucket and a full model update. The introduced modifications allow us to increase the processing speed by a factor of two or more at no or negligible compression ratio deterioration. Our algorithm limits worst-case local and global data expansion and has strictly bounded memory requirements. We present the experimental results of introduced modifications and the comparison to some well-know algorithms.

Słowa kluczowe

EN

lossless image compression; image coding; adaptive algorithms; O(n); statistical compression

• Wydawca: Institute of Information Technology of the Warsaw University of Life Sciences – SGGW
• Czasopismo: Machine Graphics and Vision
• Rocznik: 1999
• Tom: Vol. 8, No. 1
• Strony: 95--116
• Opis fizyczny: Bibliogr. 11 poz.

Twórcy

autor

Starosolski R.

• Institute of Computer Science, Silesia University of Technology,