Granular Data Compression and Representation

Abstract

We develop a comprehensive and original methodology of data compression realized in the setting of Granular Computing. It is advocated that a compression process is inherently associated with the emergence of information granules forming compressed data. This entails that compression goes hand-in-hand with the elevated level of abstraction of the generated results. The performance of the method is evaluated with the aid of the indexes of coverage and specificity commonly encountered when processing and describing information granules. A two-phase design environment is systematically established along with the detailed algorithmic layer exploring mechanisms of fuzzy clustering and the principle of justifiable granularity and its generalizations. Reconstruction error and granular reconstruction error criteria are introduced and analyzed. Experimental studies carried out on publicly available data are reported and illustrate the process of granular compression and analyze the performance of the obtained results.