LZB: Data Compression with Bounded References

Abstract

In this paper, we propose a new compression/decompression algorithm called LZB which belongs to a class of algorithms related to Lempel-Ziv (LZ). The distinguishing characteristic of LZB is that it allows decompression from arbitrary points of compressed data. This is accomplished by setting a limit on how far back a reference in compressed data can directly or indirectly point to. We enforce this limit by using a sliding "gate." During the compression, we keep track of the origin of each input symbol.  The origin of a symbol is the earliest symbol in the input data that the symbol (directly or indirectly) refers to. By using this information we avoid using any reference which go beyond the gate boundary.  We modified the gzip implementation of LZ77 to implement LZB. We then compared LZB with the alternative method in which data is segmented into smaller pieces and each piece is compressed separately by using the standard gzip. The results show that LZB improves the compression ratio by 10 to 50 percent for 1024 to 128 byte segment sizes.