Parallel Algorithms for Entropy-Coding Techniques

With the explosion of imaging applications, and due to the massive amounts of imagery data, data compression is essential. Lossless compression, also called entropy coding, is of special importance because not only it serves as a stand-alone system for certain applications such as medical imaging, it also is an inherent part of lossy compression. Therefore, fast entropy coding/decoding algorithms are desirable. In this paper we will develop parallel algorithms for several widely used entropy coding techniques, namely, arithmetic coding, run-length encoding (RLE), and Hu man coding. Our parallel arithmetic coding algorithm takes O(log N) time on an N -processor hypercube, where N is the input size. For RLE, our parallel coding and decoding algorithms take O(logN) time on N processors. Finally, in the case of Hu man coding, the parallel coding algorithm takes O(logN+n logn), where n is the alphabet size, n << N . As for decoding, however, both arithmetic and Hu man decoding are hard to parallelize. However, special provisions could be made in many applications to make arithmetic decoding and Hu man decoding fairly parallel.