Adapting the Knuth-Morris-Pratt algorithm for pattern matching in Huffman encoded texts

TLDR: A bitwise KMP algorithm is proposed that can move one extra bit in the case of a mismatch since the alphabet is binary, and two practical Huffman decoding schemes which handle more than a single bit per machine operation are combined.

Abstract: In the present work we perform compressed pattern matching in binary Huffman encoded texts [Huffman, D. (1952). A method for the construction of minimum redundancy codes, Proc. of the IRE, 40, 1098-1101]. A modified Knuth-Morris-Pratt algorithm is used in order to overcome the problem of false matches, i.e., an occurrence of the encoded pattern in the encoded text that does not correspond to an occurrence of the pattern itself in the original text. We propose a bitwise KMP algorithm that can move one extra bit in the case of a mismatch since the alphabet is binary. To avoid processing any bit of the encoded text more than once, a preprocessed table is used to determine how far to back up when a mismatch is detected, and is defined so that we are always able to align the start of the encoded pattern with the start of a codeword in the encoded text. We combine our KMP algorithm with two practical Huffman decoding schemes which handle more than a single bit per machine operation; skeleton trees defined by Klein [Klein, S. T. (2000). Skeleton trees for efficient decoding of huffman encoded texts. Information Retrieval, 3, 7-23], and numerical comparisons between special canonical values and portions of a sliding window presented in Moffat and Turpin [Moffat, A., & Turpin, A. (1997). On the implementation of minimum redundancy prefix codes. IEEE Transactions on Communications, 45, 1200-1207]. Experiments show rapid search times of our algorithms compared to the "decompress then search" method, therefore, files can be kept in their compressed form, saving memory space. When compression gain is important, these algorithms are better than cgrep [Ferragina, P., Tommasi, A., & Manzini, G. (2004). C Library to search over compressed texts, http://roquefort.di.unipi.it/~ferrax/CompressedSearch], which is only slightly faster than ours.

Full text

Adapting the Knuth-Morris-Pratt Algorithm for Pattern
Matching in Huffman Encoded Texts
Ajay Daptardar and Dana Shapira
{amax/shapird}@cs.brandeis.edu
Computer Science Department, Brandeis University, Waltham, MA
We perform compressed pattern matching in Huffman encoded texts. A modified
Knuth-Morris-Pratt (KMP) algorithm is used in order to overcome the problem of
false matches, i.e., an occurrence of the encoded pattern in the encoded text that does
not correspond to an occurrence of the pattern itself in the original text. We propose
a bitwise KMP algorithm that can move one extra bit in the case of a mismatch,
since the alphabet is binary. To avoid processing any encoded text bit more than
once, a preprocessed table is used to determine how far to back up when a mismatch
is detected, and is defined so that the encoded pattern is always aligned with the
start of a codeword in the encoded text. We combine our KMP algorithm with
two Huffman decoding algorithms which handle more than a single bit per machine
operation; Skeleton trees defined by Klein [1], and numerical comparisons between
special canonical values and portions of a sliding window presented in Moffat and
Turpin [3]. We call the combined algorithms sk-kmp and win-kmp resp e ctively.
The following table compares our algorithms with cgrep of Moura et al. [2] and
agrep which searches the uncompressed text. Columns three and four compare the
compression performance (size of the compressed text as a percentage of the uncom-
pressed text) of the Huffman code (huff ) with cgrep. The next columns compare
the processing time of pattern matching of these algorithms. The “decompress and
search” methods, which decode using skeleton trees or Moffat and Turpin’s sliding
window and search in parallel using agrep, are called sk-d and win-d respectively. The
search times are average values for patterns ranging from infrequent to frequent ones.
Files Size (bytes) Compression Search Times (sec)
cgrep huff cgrep sk-kmp win-kmp sk-d win-d
world192.txt 2,473,400 50.88 32.20 0.07 0.13 0.08 0.21 0.13
bible.txt 4,047,392 49.70 26.18 0.05 0.22 0.13 0.36 0.22
books.txt 12,582,090 52.10 30.30 0.21 0.69 0.39 1.21 0.74
95-03-erp.txt 23,976,547 34.49 25.14 0.18 1.10 0.65 1.80 1.11
As can be seen, the KMP variants are faster than the methods corresponding to
“decompress and search” but slower than cgrep. However, when compression perfor-
mance is important or when one does not want to re-compress Huffman encoded files
in order to use cgrep, the proposed algorithms are the better choice.
References
[1] Klein S.T., Skeleton Trees for efficient decoding of Huffman encoded texts, Infor-
mation Retrieval , 3, 7-23, 2000.
[2] Moura E.S., Navarro G., Ziviani N. and Baeza-Yates R., Fast and flexible
word searching on compressed Text, ACM TOIS, 18(2), 113-139, 2000.
[3] Turpin A., Moffat A., Fast file search using text compression, 20th Proc. Aus-
tralian Computer Science Conference, 1-8, 1997.