CA2686618A1

CA2686618A1 - Method and apparatus for windowing in entropy encoding

Info

Publication number: CA2686618A1
Application number: CA002686618A
Authority: CA
Inventors: Kiem-Phong Vo; Binh Dao Vo
Original assignee: At&T Corp.; Kiem-Phong Vo; Binh Dao Vo
Current assignee: AT&T Corp
Priority date: 2003-07-17
Filing date: 2004-07-19
Publication date: 2005-01-17
Also published as: CA2475186A1; CA2475186C; CA2475189A1; CA2475189C; US20110173167A1; US20050055367A1; US20050044294A1; US7925639B2; US8200680B2; US20080040375A1; US7296030B2; US7454431B2

Abstract

The present invention provides efficient window partitioning algorithms for entropy-encoding. The present invention enhances compression performance of entropy encoding based on the approach of modeling a dataset with the frequencies of its n-grams. The present invention may then employ approximation algorithms to compute good partitions in time O(s*log s) and O(s) respectively, for any data segment S with length s.

Description

METHOD AND APPARATUS FOR WINDOWING IN ENTROPY ENCODING

This is a divisional application of Canadian Patent Application 2,475,186 filed July 19, 2004.

[0001] The present invention relates generally to data compression and, more particularly, to a method for efficient window partition identification in entropy encoding to enhance compression performance of entropy encoding based on the idea of modeling a dataset with the frequencies of its n-grams.

BACKGROUND OF THE INVENTION

[0002] Compression programs routinely limit the data to be compressed together in segments called windows. The process of doing this is called windowing.
String-based compression techniques such as Lempel-Ziv or Burrows-Wheeler often use fixed-size windows suitable for in-core processing. Entropy-encoding techniques such as Huffman or arithmetic compression normally do not require windowing except to bound code lengths or to avoid reading large files multiple times.
However, these compressors can benefit from windowing when the statistical models change in different file regions. For example, consider a data file made up from four letters in which two letters appear exclusively in the first half of the file while the other two letters appear exclusively in the second half. If all letters appear with the same frequency, a Huffman compressor would normally encode each letter with two bits.
On the other hand, each letter can be encoded with a single bit if each half of the file is treated separately. Adaptive techniques such as adaptive Huffman or splay tr-ee do encode data with shifting models but they often produce inferior codes and incur larger costs in both compression and uncompression times than static Huffman.

[0003] Therefore, a need exists for a method for efficient window partition identification in entropy encoding, e.g., with performance much better than O(s) time.
SUMMARY OF THE INVENTION

[0004] In one embodiment, the present invention significantly improves the performance of identifying window partitions in entropy encoding. In particular, the -la-present invention, enhances compression performance of entropy encoding based on the idea of modeling a dataset with the frequencies of its n-grams and employs two approximation algorithms to compute good partitions in time O(s*Iog s) and O(s) respectively, for any data segment S with length s.

[0005] Certain exemplary embodiments may provide a computer implemented method for partitioning a given data segment, P, with length p into smaller segments, P;, that can be compressed separately, said method comprising the steps of: (a) receiving from a storage medium a data segment P;
(b) processing said data segment P using a processor, the processing comprising:
(b1) dividing said data segment, P into a plurality of 2-partition pairs a where both partitions within each pair having a predefined length of at least (P); (b2) estimating a compressed length of each partition within each pair; (b3) selecting one of said 2-partition pairs 7r for partitioning said data segment, P, such that r(71)<
F(P), where I'(7c) represents a compressed length of said partition and F(P) represents a compressed length of said data segment, P; and (c) storing a selected one of said 2-partition pairs 7c.

[0005a] Certain other exemplary embodiments may provide a computer-readable medium having stored thereon a plurality of instructions, the plurality of instructions including instructions which, when executed by a processor, cause the processor to perform the steps of a method for partitioning a given data segment, P, with length p into smaller segments, P;, that can be compressed separately, comprising of: dividing said data segment, P into a plurality of 2-partition pairs 7c where both partitions within each pair having a predefined length of at least (P); estimating the compressed length of each partition within each pair; and selecting one of said 2-partition pairs 71 for partitioning said data segment, P, such that I'(7c)< I'(P), where r'(7E) represents a compressed length of said partition and F(P) represents a compressed length of said data segment, P.

-lb-[0005b] Still certain other exemplary embodiments may provide an apparatus for partitioning a given data segment, P, with length p into smaller segments, P;, that can be compressed separately, comprising: a processor for dividing said data segment, P into a plurality of 2-partition pairs n where both partitions within each pair having a predefined length of at least (P), estimating the compressed length of each partition within each pair, and selecting one of said 2-partition pairs n for partitioning said data segment, P, such that I'(7[)<
F(P), where I'(7c) represents a compressed length of said partition and r(P) represents a compressed length of said data segment, P.

BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The teaching of the present invention can be readily understood by considering the following detailed description in conjunction with the accompanying drawings, in which:

[0007] FIG. 1 illustrates a flowchart of a partitioning method of the present invention for recursively partitioning a given data segment into smalL-r segments that can be compressed separately;

[0008] FIG. 2 illustrates an example of all possible 2-partitions of a data segment P, having length p, with parts having length at least (P) long;

[0009] FIG. 3 illustrates a flowchart of a faster variation of partitioning method of the present invention for recursively partitioning a given data segment into smaller segments that can be compressed separately; and [0010] FIG. 4 illustrates the present partitioning method implemented using a general purpose computer or any other hardware equivalents.

[0011] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures.
DETAILED DESCRIPTION

[0012] The present invention relates to data compression using entropy encoding based on the idea of modeling a dataset with the frequencies of its n-grams.

[0013] To better understand the present invention, a description of n-grams and its use are first provided. The present invention uses n-grams to model data.
For any data segment S with length s, an n-gram in S is a subsequence of n < s consecutive bytes. Assume an arbitrary but fixed n, the notation Si will denote the n-gram in S starting at location i while S[i] denotes the byte at location i.
For example, the string S = abababac is of length 8 and has five 4-grams, of which three are distinct: abab, baba and abac. The 4-grams So and S2 are the same: abab.

[0014] In one embodiment, the present invention repeatedly examine n-gram frequencies of given data segments. Thus, it is beneficial if this step can be executed quickly. For any general data segment S, the notation Fs shall be used to denote an associative array of frequencies indexed by the n-grams of S.
Suppose that Fs was initialized to 0's, the below loop computes all n-gram frequencies:

for(i=0; i<=s-n; i+=1) Fs[Si]+=1;

[0015] This loop runs in time O(s) as long as the cost of indexing the array Fs can be bounded by some constant. This can be ensured by implementing Fs as a hash table indexed by the distinct n-grams. However, hash table look-up cost is significant and the frequency estimates do not always need to be exact. Thus, Fs is chosen to be implemented as a normal array of size A by hashing each n-gram Si to an integer via the below hash function with some preselected constant a:

x(Si) =(an-1S[i] + an-2S[i] + ... + S[i+n-1]) mod A (Equ. 1) The above loop then becomes:
for(i=0; i<=s-n; i+=1) Fs[x(S;)]+=1;

[0016] For nontrivial values of n, the loop can be further optimized by exploiting the linearity of the hash function x to compute x(Si+,) from x(S;) via:

x(S;+l) ={a(x(S;) - a"-lS[i]} + S[i+n]} mod A (Equ. 2) [0017] Computing frequencies of 1-grams or single letters is, of course, the basis for Huffman and arithmetic coders. For entropy-encoding compressors, n=1 so A=256 would allow FS to index all possible 8-bit bytes at no loss of accuracy.
Henceforth, given a data segment S and a frequency array FS, it shall be assumed that FS is indexed by mapping the n-grams via the x function as described.
Therefore, the notation FS[S;] will mean Fs[x(Si)].

[0018] Entropy-encoding compressors such as Huffman and arithmetic coders compress data based on modeling the probability of symbols which would be 1-grams or bytes. Although these compressors are sensitive to changes in symbol statistics, they often cannot adapt to evolution in the statistical models.
Certain adaptive versions of these algorithms can cope with some model changes but tend to produce less efficient codes and have longer running time. Overall, none of these schemes work well when the statistical models abruptly change. For example, Buchsbaum et al. developed a dynamic programming solution to the problem of grouping columns in large tables to enhance compression. In entropy encoding application with any data segment S with s bytes in length, since each byte is treated as a column, the algorithm can be used to compute an optimal partition in O(s) time.
This is too slow for large datasets with size in the megabytes.

[0019] To address this criticality, the present invention provides two methods for computing good partitions using approximations, with significant performance enhancements, in O(s*Iog s) and O(s) time respectively. This also means that there is good potential gain in finding a good partition of the data into sections with sufficiently different symbol statistics using the present invention, then applying the compressors to each section separately.

[0020] Let S be a data segment of length s. A partition n of S with k parts divides S into a sequence of consecutive non-overlapping sub-segments (P,, P2, ... Pk) that together exactly cover S. A partition with k parts is referred as a k-partition. For a given compressor F and a data segment P, let F(P) be the compressed length of P.
Then, the compressed length of any partition n of S with k parts is:

k I'(7C) = L T(P;) (Equ. 3) i=1 [0021] FIG. 1 illustrates a flowchart of a partition method 100 of the present invention for recursively partitioning a given data segment into smaller segments that can be compressed separately computation complexity of O(slog s) time. Method 100 starts in step 105 and proceeds to step 110.

[0022] In step 110, a data segment P is accepted as input to be partitioned.
In step 120, the length of segment P, p, is checked if it is smaller than 2* (P).
If p is smaller than 2* (P), then the method terminates in step 160; otherwise, the method proceeds to step 130.

[0023] In step 130, among all 2-partitions of P with both partitions having length at least (P), find an such that r(n) < I'(P). Ingeneral, for a segment P with length p with parts having length at least (P) in length, there will be a total of (p-2 (P)+1) 2-partitions. FIG. 2 illustrates the possible combinations of 2-partitions for a segment P with parts having length at least (P) in length. Each 2-partition contains a left and a right partition. In the first combination, nl, the left partition have length (P) and the right partition have length p - (P). In the second combination, n2, the left partition have length (P)+1 and the right partition have length p - (P)-1.
In the third combination, n3, the left partition have length (P)+2 and the right partition have length p - (P)-2. Following this pattern, in the last combination, nP_2 (p)+1, which is the (p-2 (P)+1)th combination, the left partition have length p - (P) and the right partition have length (P). Therefore, in step 130, among all possible 2-partition combinations for a segment P with parts having length at least (P) in length, the k method calculates F(n) =L I,(pi) based on Equ. 3 and check if the condition r(n) <
i=1 I'(P) is true for each 2-partition.

[0024] In step 140, if a 2-partition that meets the condition of r(n) < I'(P), then the method proceeds to step 150. It is possible that more than one 2-partitions, 7t, can be found to meet such condition. In that case, the choice of a good n is arbitrary and depends on the applications of the algorithm. In one embodiment of the present invention, the good it used can simply be the first 7t found among all 2-partitions. If a 2-partition cannot be found to meet the condition of r(n) < r(P) among all 2-partitions, then the method terminates in step 160.

[0025] In step 150, for the chosen 2-partition that meet the condition of F(n) <
r(P), the method recursiveiy and independently uses the left partition and the right partition as inputs to method 100.

[0026] In one embodiment of the present invention, (P) is chosen to be equal to max(p/5, c), where p is the length of P and E=212, i.e. 4K bytes. This function is used to determine the minimum length of a partition since small data segments compress poorly. By requiring (P) to be a fixed fraction of the length of P, the depth of the recursion of method 100 can be bounded by 0(log s). Then, the entire algorithm runs in O(yslog s) time where s is the length of the original data segment S
and 7 is an estimation of the cost to compute the compressed length function F on the parts of the candidate partitions.

[0027] For a general compressor, the only way to compute I' might be to invoke the cornpressor on the data itself and measure the result. In that case y might be up to 0(s). For entropy-encoding compressors, it is possible to define an estimation function with constant time amortized cost. Consider a data segment P of length p at any recursion level in method 100. Let Fp the corresponding array of byte frequencies. Shannon's information theory asserts that the number of bits required to encode a byte i with respect to the data in P is log(p/ FP{i]) since Fp[i]/p is the empirical probability of i. Let i be an estimate for the length of a table of codes or frequencies that a static Huffman or arithmetic compressor would need to decode data. Then, the compressed length of P, I'e(P), can be estimated with:

Fe(P) = i + I Fp[i]io9( p ) ='r + piogp - I Fp[i]io9(Fp[i]) (Equ. 4) i=o Fp [i] i=O

[0028] In one embodiment of the present invention, ti= 5b + (256-b) where b is the number of bytes with non-zero frequency. The factor of 5 was chosen because the Huffman encoder used in one embodiment of the present invention guarantees maximal code length 32. The term 256 - b estimates the space needed to encode the bytes not appearing in the data, i.e., having zero code length.

[0029] Now, let itl =(Pl, P2) and n2 =(Ql, Q2) be two 2-partitions of P such that Q, is formed by extending P, by one byte on the right. Then Q2 must have been formed by cutting one byte from the left of P2. Since only a single byte ieaves a part or gets added to it, the frequency arrays FP1 and FP7 , can be updated in constant time to form FQ, and FQz . As a consequence, re(n2) can be computed in constant time from I'e(7E,).

[0030] Since all 2-partitions of can be generated by exchanging bytes in a loop starting from the partition (0, P), where 0 is a null data segment, step 130 of method 100 can be implemented so that the total running time of all invocations of the compressed iength function Fe is O(p). Thus, the amortized cost of each Fe is constant. Further, since each recursion level of method 100 only needs two frequency arrays in the computing loop, the required space for method 100 at all recursion levels is bounded by O(Iog s). Putting everything together, method can be implemented in O(slog s) time and O(s + logs) space where s is the length of the data segment to be partitioned.

[0031] For a slight loss in compression performance, it is possible to eliminate the factor log s from the time complexity of method 100. FIG. 3 illustrates a flowchart of a faster partition method 300 of the present invention for recursively partitioning a given data segment into smaller segments that can be compressed separately with computation complexity of O(s) time. Method 300 starts in step 305 and proceeds to step 310.

[0032] In step 310, a data segment P is accepted as input to be partitioned.
In step 320, the length of segment P, p, is checked if it is smaller than 2* (P).
If p is smaller than 2p(P), then the method terminates in step 380; otherwise, the method proceeds to step 330.

[0033] In step 330, all 2-partitions with parts having minimum length of (P) are first ordered by the length of their left parts. FIG. 2 illustrates an example of the outcome of such an ordering step. In general, for a segment P with length p with parts having length at least p(P) in length, there will be a total of (p-2 (P)+1) 2-partitions. In the first 2-partition, nl, the left partition have length (P) and the right partition have length p - (P). In the second 2-partition, 712, the left partition have length (P)+1 and the right partition have length p - p(P)-1. In the third 2-partition, n3, the left partition have length p(P)+2 and the right partition have length p - (P)-2.
Following this pattern, in the last 2-partition, 7ip_2N(p)+1, which is the (p-2 (P)+1)th 2-partition, the left partition have length p - (P) and the right partition have length (P).
Then, step 330 initializes the variables i to 1 and N to p-2 (P).

[0034] In step 340, if i is greater than N, then the method terminates in step 380;
otherwise, the method proceeds to step 350. In step 350, if I'(ui) < r(P) and T'(n;,) >
I'(7r;), then the method proceeds to step 370; otherwise, the method proceeds to step 360. In step 360, the method increments i6y 1 and the proceeds back to step 340.
In step 370, the method recursively apply the right partition of zc; as input to method 300.

[0035] The basic idea behind method 300 is to consider all 2-partitions of S
in order starting from (0, S), where 0 is a null data segment. When a partition is found that improves over the encoding of the entire data segment, it is simply split off from its left part, then used to iterate on the rest. The machinery developed earlier to update frequency arrays can be applied straightforwardly here so that method can be implemented in O(s) time and space where s is the length of the data segment to be partitioned.

[0036] FIG. 4 illustrates the present partitioning method implemented using a general purpose computer 400 or any other hardware equivalents. For example, the present partitioning methods and data structures can be represented by one or more software applications 405 (or even a combination of software and hardware, e.g., using application specific integrated circuits (ASIC)), where the software is loaded from a storage medium 406, (e.g., a ROM, a magnetic or optical drive or diskette) and operated by the CPU 402 in the memory 404 of the system. As such, the present partitioning methods and data structures of the present invention can be stored on a computer readable medium, e.g., RAM memory, ROM, magnetic or optical drive or diskette and the like.

[0037] While various embodiments have been described above, it should be understood that they have been presented by way of example only, and not limitation. Thus, the breadth and scope of a preferred embodiment should not be limited by any of the above-described exemplary embodiments, but shouid be defined only in accordance with the following claims and their equivalents.

Claims

1. A computer implemented method for partitioning a given data segment, P, with length p into smaller segments, P i, that can be compressed separately, said method comprising the steps of:
(a) receiving from a storage medium a data segment P;
(b) processing said data segment P using a processor, the processing comprising:
(b1) dividing said data segment, P into a plurality of 2-partition pairs .pi.
where both partitions within each pair having a predefined length of at least µ(P);
(b2) estimating a compressed length of each partition within each pair;
(b3) selecting one of said 2-partition pairs n for partitioning said data segment, P, such that .GAMMA.(.pi.)< .GAMMA.(P), where .GAMMA.(.pi.) represents a compressed length of said partition and .GAMMA.(P) represents a compressed length of said data segment, P; and (c) storing a selected one of said 2-partition pairs .pi..

2. The method of claim 1, wherein if more than one 2-partition pair .pi. is determined to satisfy .GAMMA.(.pi.) < .GAMMA.(P), then said dividing step, said estimating step and said selecting step is recursively applied to each partition of said more than one 2-partition pair n.

3. The method of claim 1, wherein said the compressed length of said data segment is estimated using the function of:

where F P[i] is a corresponding array of frequencies for a byte i and .tau. is an estimate for the length of a table of codes or frequencies that a static Huffman or arithmetic compressor needs to decode data.

4. The method of claim 1, wherein said method has a computation complexity O(plog p).

5. The method of claim 1, where and µ(P)=max(p/5, .epsilon.) and .epsilon. is an arbitrary parameter.

6. The method of claim 2, further comprising:
picking any one of said .pi. arbitrarily when more than one instance of said .pi. is found.

7. The method of claim 1, further comprising:
ordering all 2-partition pairs by the length of their left parts, for all 2-partitions of P with parts having minimum length of p(P), where µ(P)=max(p/5, .epsilon.) and E is an arbitrary parameter; and finding the first .pi. i such that .GAMMA.(.pi. i) < .GAMMA.(P) and .GAMMA.(.pi. i+1) > .GAMMA.(.pi.i).

8. The method of claim 7, wherein said finding step comprises:
when said .pi. i is found, then recursively applying a right partition associated with said .pi. i.

9. A computer-readable medium having stored thereon a plurality of instructions, the plurality of instructions including instructions which, when executed by a processor, cause the processor to perform the steps of a method for partitioning a given data segment, P, with length p into smaller segments, P i, that can be compressed separately, comprising of:
dividing said data segment, P into a plurality of 2-partition pairs n where both partitions within each pair having a predefined length of at least µ(P);
estimating the compressed length of each partition within each pair; and selecting one of said 2-partition pairs .pi. for partitioning said data segment, P, such that .GAMMA.(.pi.) < .GAMMA.(P), where .GAMMA.(.pi.) represents a compressed length of said partition and .GAMMA.(P) represents a compressed length of said data segment, P.

10. The computer-readable medium of claim 9, wherein if more than one 2-partition pair .pi. is determined to satisfy .GAMMA.(.pi.) < .GAMMA.(P), then said dividing step, said estimating step and said selecting step is recursively applied to each partition of said more than one 2-partition pair .pi..

11. The computer-readable medium of claim 9, wherein said the compressed length of said data segment is estimated using the function of:

where F p[i] is a corresponding array of frequencies for a byte i and .tau. is an estimate for the length of a table of codes or frequencies that a static Huffman or arithmetic compressor needs to decode data.

12. The computer-readable medium of claim 9, wherein said method has a computation complexity O(plog p).

13. The computer-readable medium of claim 9, where and µ(P)=max(p/5, .epsilon.) and .epsilon. is an arbitrary parameter.

14. The computer-readable medium of claim 10, further comprising:
picking any one of said n arbitrarily when more than one instance of said .pi.
is found.

15. The computer-readable medium of claim 9, further comprising:
ordering all 2-partition pairs by the length of their left parts, for all 2-partitions of P with parts having minimum length of µ(P), where µ(P)=max(p/5, .epsilon.) and .epsilon. is an arbitrary parameter; and finding the first .pi. i such that .GAMMA.(.pi. i) < .GAMMA.(P) and .GAMMA.(.pi. i+1) > .GAMMA.(.pi. i).

16. The computer-readable medium of claim 15, wherein said finding step comprises:
when said .pi.i is found, then recursively applying a right partition associated with said .pi.i.

17. An apparatus for partitioning a given data segment, P, with length p into smaller segments, P i, that can be compressed separately, comprising:
a processor for dividing said data segment, P into a plurality of 2-partition pairs .pi. where both partitions within each pair having a predefined length of at least µ(P) estimating the compressed length of each partition within each pair, and selecting one of said 2-partition pairs .pi. for partitioning said data segment, P, such that .GAMMA.(.pi.)< .GAMMA.(P), where .GAMMA.(.pi.) represents a compressed length of said partition and .GAMMA.(P) represents a compressed length of said data segment, P.

18. The apparatus of claim 17, wherein if more than one 2-partition pair n is determined to satisfy .GAMMA.(.pi.) < .GAMMA.(P), then said dividing step, said estimating step and said selecting step is recursively applied to each partition of said more than one 2-partition pair .pi..

19. The apparatus of claim 17, wherein said the compressed length of said data segment is estimated using the function of:

where F P[i] is a corresponding array of frequencies for a byte i and .tau. is an estimate for the length of a table of codes or frequencies that a static Huffman or arithmetic compressor needs to decode data.

20. The apparatus of claim 17, wherein said method has a computation complexity O(plog p).