Bitmaps are a useful, but storage voracious, component of many information retrieval systems. Earlier efforts to compress bitmaps were based on models of bit generation, particularly Markov models. While these permitted considerable reduction in storage, the short memory of Markov models may limit their compression efficiency. In this paper we accept the state orientation of Markov models, but introduce a Bayesian approach to assess the state; the analysis is based on data accumulating in a growing window. The paper describes the details of the probabilistic assumptions governing the Bayesian analysis, as well as the protocol for controlling the window that receives the data. We find slight improvement over the best performing strictly Markov models.
Bibliographical noteFunding Information:
This work was supported, in part, by NSF Grant IRI-9307895-A01 (A.B.), grant 8560195 of the Israeli Ministry of Science (S.K.) and grant 865431 of the Academy of Finland (T.R.). The authors gratefully acknowledge these supports.
- Bitmap compression
- IR models
- Markov modelling