Efficient Integer Retrieval from Unordered Compressed Sequences
Abstract: |
We investigate the problem of fast direct access to elements of an integer sequence given in a compressed form. If integers are sorted in ascending order, it can be reduced to performing the 'select' operation on a bitmap, which is very well investigated. We focus on the more general and more complicated case of unordered integer sequence and propose to represent it with the help of variable-length Reverse Multi-Delimiter (RMD) codes. When applied to data compression, these codes combine a good compression ratio with fast decoding. In this paper, another property of RMD codes is researched - the ability of direct access to codewords in the encoded bitstream. We present the method allowing us to extract and decode a codeword from an RMD-bitstream in almost constant time and experiment on its application to natural language text compression. Due to the properties of RMD codes and compactness of auxiliary direct access structures, our method appears to be very space efficient, only a few percent above the Shannon entropy. |
Download paper: | |||
PostScript | BibTeX reference |