P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3-12, Snowbird 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and contract 1419991431A from sponsors of CERIAS at Purdue. consists in determining for a class S of source models the growth rate of n (S) min fR n (C n ; P )g; 1) R n (S) min fR n (C n ; P )g: 2) In this paper, we investigate the average redundancy of arithmetic coding [6] for memoryless sources with unknown parameters. Here, we analyze an idealized arithmetic coding in which nite precision and nite bu er sizes are not taken into account. Following [16] we assume that the idealized arithmetic encoding consists of the Krichevsky and Tro mov estimator followed by the ....

....with k 1 and n k 0 , that is, P (x . It is assumed that is unknown. Therefore, to estimate the probability P (x 1 ) we shall use the KT estimator [7, 16] de ned as P e (k; n k) k 1=2) n k 1=2) n) To generate an arithmetic encoding, we apply the Shannon Fano code (cf. [2, 6]) for the probability distribution P e (k; n k) That is, the code length L n is L n = d log P e (k; n k)e 1. The average redundancy of the arithmetic coding therefore becomes d log P e (k; n k)e log Using dxe = x 1 hxi, where hxi is the fractional part of x, we reduce the ....

P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3-12, Snowbird 1991.

....of the objects in the image, their encoding may not be the most efficient. Even with DPCM coding followed by arithmetic 92 coding for the DVF, too large a window size and thus too large a potential maximal value would dilute the probability model and thus yield a sub optimal coding rate. In [31] the authors showed that using adaptive arithmetic coding, the description length, R, of a source of alphabet size n, is R = log 2 B B B (n i) C C C (8.1) where k is the number of alphabet symbols that occur in the stream to be compressed, c i is the number of occurrences of ....

P. G. Howard and J. S. Vitter. Analysis of arithmetic coding for data compression. In Proceedings of the Data Compression Conference, volume 1, pages 3--12, April 1991.

.... N Gamma 1g is the set of preimages ff Gamma1 (l) l 2 L(T )g: The cost c(d 0 ; d 1 ) for d 0 ; d 1 2 Z 0 is i=0 (i ffi) j=0 (j ffi) Q d 0 d 1 Gamma1 i=0 (i 2ffi) This definition corresponds, algorithmically, to the use of a one pass arithmetic coder without modelling [1] . The time required to do one pass arithmetic coding on binary data is linear, and, hence, will be disregarded in our proof of linearity. The time required to compute c(d 0 ; d 1 ) is constant, if one uses a table for small values of d 0 ; d 1 and Stirling s approximation for values that are ....

Howard, Paul G., and Jeffrey Scott Vitter, "Analysis of Arithmetic Coding for Data Compression", in Proceedings of the

....we want this compression method to be suitable for web based applications, the instruction parameters are multiplexed in the same arithmetic coding stream, so that the image video can be displayed on the y while downloading. We implement our own arithmetic encoder based on the ideas discussed in [12], hence we refrain from describing it in detail. B. Distortion Measures The process of lossy compression can be viewed as adding noise to the image. For this reason, we also refer to the distortion level D as RandomNoise. The original pattern matching scheme PMIC [3] computed the ....

P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3-12, Snowbird 1991.

....0 s and other small integers H 0 (s) will be small and s will be efficiently compressed by a zeroth order algorithm. Note that neither Huffman coding nor arithmetic coding are able to achieve compression ratio H 0 (s) for every string s. However, arithmetic coding can get quite close to that: in [15] Howard and Vitter proved that the arithmetic coding procedure described in [38] is such that for every string s its output size Arit(s) is bounded by Arit(s) jsjH 0 (s) 1 jsj 2 (2) with 1 10 Gamma2 . In other words, the compression ratio Arit(s) jsj is bounded by the entropy plus a ....

P. Howard and J. Vitter. Analysis of arithmetic coding for data compression. Information Processing and Management, 28(6), 1992.

....to near mH(p) bits with high probability, with the deviation from the average having a Cherno# like bound. For more information on arithmetic coding, we refer the reader to [9, 11] For more precise statements and details regarding the low variability of arithmetic coding, we refer the reader to [7]. More details will appear in the final version. Given this compression scheme, we suggest the following approach. Choose a maximum desired uncompressed size m. Then design a compressed Bloom filter using the above theory using a slightly smaller compressed size than desired; for example, if the ....

P. G. Howard and J. Vitter. Analysis of arithmetic coding for data compression. Information Processing and Management, vol 28. no. 6, pages 749-763, 1992.

....of integers N and pN ) but that extra cost is bounded by 3dlog 2 Ne = o(N) bits. However, using exact arithmetic one cannot compute the interval I N in linear time as the numbers used to represent the endpoints of I j s grow in size with each step of the algorithm. A way out of this problem (see [6] or [16] for a discussion) is to approximate each of the intervals I j with a subinterval of [0; 1] whose endpoints are quotients with equal denominators being powers of 2 and numerators are B bit integers di ering by more than 2 B 2 . This allows to compute I j from I j 1 by means of a nite ....

....integers, but, because of the necessity of rounding the split point causes the ratio of the lengths of I j 1 and I j to be slightly di erent from p or 1 p (as in the idealistic variant) the di erence being bounded by 2 2 B . A straighforward calculation being a special case of the results of [6] shows that the extra encoding cost caused by rounding can be bounded by 2 2 B (B 2) 2 3 B bits per triangle. This means that the bound on the length of the entire encoding is M(f) 2 2 B (B 2) 2 3 B 3dlog 2 te=t bits per triangle (where t stands for the number of triangles) ....

P.G.Howard and J.S.Vitter. Analysis of Arithmetic Coding for Data Compression, Information Processing and Management, 28(6), 1992, 749-763.

....x 0 = x 1 , x 2 , x c such that for all j 1 substring x j without its last character is equal to some x i , for 0 i j. It encodes the substring x j by the value i, using dlg je bits, followed by the ascii encoding of the last character of x j , using dlg ffe bits. Arithmetic coding [HoV, Lanb, WNC] is a coding technique that achieves a coding length equal to the entropy of the data model. Sequences of probability p are encoded using lg(1=p) bits. Arithmetic coding can be thought of as using fractional bits, as opposed to the suboptimal Huffman coding in which all code lengths must be ....

....E by building a probabilistic model that feeds probability information to an arithmetic coder [BCW, Lana] as explained in the example below. It has been shown that the coding length obtained in this character based approach is at least as good as that obtained using the word based approach [BCW, HoV, Lana]. Hence, the optimality results in [ZiLb] hold without change for the character based approach. Example 3.1 Assume for simplicity that our alphabet is fa; bg. We consider the page request sequence aaaababaabbbabaa: The Ziv Lempel encoder parses this string as ....

[Article contains additional citation context not shown here]

P. G. Howard and J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression, " Information Processing and Management 28 (1992), 749--763, invited paper in Special Issue on Data Compression for Images and Texts.

....Huffman coding (11) holds with = 1. For the dynamic Huffman coding algorithm described in [22] 11) holds with = 2. Arithmetic coding routines exist in different flavors (see for example [10, 15, 25] each one with a different balance between storage requirements, compression, and speed. In [9] Howard and Vitter carry out a comprehensive analysis of arithmetic coding which tells us that a simple arithmetic coder, such as the one described in [25] satisfies (11) with 10 Gamma2 . 4 Analysis of the algorithm BW0 In this section we analyze the algorithm BW0 = bwt mtf Order0, ....

P. Howard and J. Vitter. Analysis of arithmetic coding for data compression. Information Processing and Management, 28(6), 1992.

....compression method to be suitable for WWW based applications, the instruction parameters are multiplexed in the same arithmetic coding stream, so that the image video can be displayed on the fly while downloading. In fact, we implemented our own arithmetic encoder based on the ideas discussed in [11], hence we refrain from describing it in details. 3.4.4 Overall Space and Time Complexity We shall argue that the overall time complexity for the compression is the optimal O(N 2 ) for decompression O(N 2 = log N ) and the space complexity is O(N 2 ) Let us first consider the compression ....

P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3--12, Snowbird 1991.

....intervals. Subsequently we refer to this method by the name WP. The coding scheme WP falls into the category of adaptive arithmetic data compression methods (Witten, Neal Cleary, 1987) Given the class frequency distribution for a sample S of n examples, we can compute its exact code length (Howard Vitter, 1992): WP(S; ff) log 2 (mff) n Q m j=1 ff n j ; where a b is a shorthand notation for the increasing power a(a 1) Delta Delta Delta (a b Gamma 1) A natural way of defining the MDL MML cost of a k split is WP cost k ] i=1 S i = log 2 (V Gamma 1) log 2 V Gamma 1 k ....

Howard, P. G., & Vitter, J. S. (1992). Analysis of arithmetic coding for data compression.

....compression method to be suitable for WWW based applications, the instruction parameters are multiplexed in the same arithmetic coding stream, so that the image video can be displayed on the fly while downloading. In fact, we implemented our own arithmetic encoder based on the ideas discussed in [11], hence we refrain from describing it in detail. 3.2 Distortion Measures The process of lossy compression can be viewed as adding noise to the image. For this reason, we refer to distortion D as RandomNoise. The original pattern matching scheme PMC [2] computed the distortion per symbol for ....

P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3--12, Snowbird 1991.

....information theory [9] applied to predicting [8,12,28] and generalize P 1 to get a universal prefetcher P that is optimal in the limit against a general finite state prefetcher. The resulting optimal prefetcher P is a blend of P 1 and the prefetcher [36] based on the Lempel Ziv data compressor [18,25,37]. We show in Section 6 how to implement the prefetcher in constant expected time per prefetched page, independent of alphabet size ff and cache size k, by use of the optimal dynamic algorithm for generating discrete random variates of Matias, Vitter, and Ni [26] which uses a table lookup method ....

.... the convergence rate cannot be faster than O(1= n ) 7] The importance of the above theorem lies in its generalization to higher order using techniques from information theory [9] The approach of [12] allows us to combine P 1 with a prefetcher [36] based on the Lempel Ziv data compressor [18,25,37] to get a prefetcher P that is optimal in the limit against the class of finite state prefetchers. 1 of length n drawn from alphabet A, and any s 0, the fault rate of prefetcher P on oe 1 converges almost surely to Fault F(s) oe 1 ) as n 1. From the observation in Section 1.1 that ....

P. G. Howard and J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression," Information Processing and Management 28 (1992), 749--763, invited paper in Special Issue on Data Compression for Images and Texts.

....draw on ideas from information theory [9] applied to predicting [8, 12, 28] and generalize P 1 to get a universal prefetcher P that is optimal in the limit against a general FSP. The resulting optimal prefetcher P is a blend of P 1 and the prefetcher [36] based on the Lempel Ziv data compressor [18, 25, 37]. We show in section 6 how to implement the prefetcher in constant expected time per prefetched page, independent of alphabet size # and cache size k, by using the optimal dynamic algorithm for generating discrete random variates of Matias, Vitter, and Ni [26] which uses a table lookup method of ....

.... the convergence rate cannot be faster than O(1 # n ) 7] The importance of the above theorem lies in its generalization to higher order using techniques from information theory [9] The approach of [12] allows us to combine P 1 with a prefetcher [36] based on the Lempel Ziv data compressor [18, 25, 37] to get a prefetcher P that is optimal in the limit against the class of FSPs. Theorem 2. For every sequence # n 1 of length n drawn from alphabet A, and any s # 0, the fault rate of prefetcher P on # n 1 converges almost surely to Fault F(s) # n 1 ) as n ##. From the observation ....

P. G. Howard and J. S. Vitter, Analysis of arithmetic coding for data compression, Invited paper in Special Issue on Data Compression for Images and Texts, Inform. Process. Management, 28 (1992), pp. 749--763.

.... require two passes and transmission of model data as side information; if the model data is transmitted efficiently they can provide slightly better compression than adaptive codes, but in general the cost of transmitting the model is about the same as the learning cost in the adaptive case [4]. To get good compression we need models that go beyond global event counts and take into account the structure of the data. For images this usually means using the numeric intensity values of nearby pixels to predict the intensity of each new pixel and using a suitable probability distribution ....

....to scale the counts periodically, typically by halving all weights when the total weight reaches a specified number. This effectively gives more weight to more recent occurrences of each letter, and has the additional benefit of keeping the counts small enough to fit into small registers. Analysis [4] shows that scaling often helps compression, and can never hurt it by very much. One technique for probability estimation when only two events are possible is to use small scaled counts, considering each count pair to be a probability state. We can then precompute both the corresponding ....

P. G. Howard and J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression," Information Processing and Management, 28, no. 6, pp. 749--763, 1992.

....the following property: For each j 1, substring x j without its last character is equal to some previous substring x i , where 0 i j. Substring x j is encoded by the value i, using dlg je bits, followed by the ascii encoding of the last character of x j , using dlg ffe bits. Arithmetic coding [HoV, Lanb, WNC] is a coding technique that achieves a coding length equal to the entropy of the data model. Sequences of probability p are encoded using lg(1=p) Gamma lg p bits. Arithmetic coding can be thought of as using fractional bits, as opposed to the suboptimal Huffman coding in which all code ....

....E by building a probabilistic model that feeds probability information to an arithmetic coder [BCW, Lana] as explained in the example below. It has been shown that the coding length obtained in this character based approach is at least as good as that obtained using the word based approach [BCW, HoV, Lana]. Hence, the optimality results in [ZiL] hold without change for the character based approach. Example 1 Assume for simplicity that our alphabet is fa; bg. We consider the page request sequence aaaababaabbbabaa : The Ziv Lempel encoder parses this string as (a) aa) ab) aba) abb) b) abaa) ....

[Article contains additional citation context not shown here]

P. G. Howard & J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression," Information Processing and Management 28 (1992), 749--763, invited paper in Special Issue on Data Compression for Images and Texts.

....errors are made easy to detect, and upon detection of an error, bits are changed until no errors are detected. Overview of this paper. In Section 2 we give a tutorial on arithmetic coding. We include an introduction to modeling for text compression. We also restate several important theorems from [22] relating to the optimality of arithmetic coding in theory and in practice. In Section 3 we present some of our current research into practical ways of improving the speed of arithmetic coding without sacrificing much compression efficiency. The center of this research is a reduced precision ....

.... [36; 100) It is possible to maintain higher precision, truncating (and adjusting to avoid overlapping subintervals) only when the expansion process is complete; this makes it possible to prove a tight analytical bound on the lost compression caused by the use of integer arithmetic, as we do in [22], restated as Theorem 1 below. In practice this refinement makes the coding more difficult without improving compression. 2 Analysis. In [22] we prove a number of theorems about the code lengths of files coded with arithmetic coding. Most of the results involve the use of arithmetic coding in ....

[Article contains additional citation context not shown here]

P. G. Howard & J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression, " Information Processing and Management 28 (1992), 749--763.

....theory [CoT] applied to predicting [CoS, FMG, MeF] and generalize P 1 to get a universal prefetcher P that is optimal in the limit against a general finite state prefetcher. The resulting optimal prefetcher P is a blend of P 1 and the prefetcher [ViK] based on the Lempel Ziv data compressor [HoV, Lan, ZiL] (on which the UNIX compress program is based) We show in Section 6 how to implement the prefetcher in constant expected time per prefetched page, independent of alphabet size ff and cache size k, by use of the newly developed optimal dynamic algorithm for generating discrete random variates ....

.... convergence rate cannot be faster than O(1= p n ) Cov] The importance of the above theorem lies in its generalization to higher order using techniques from information theory [CoT] The approach of [FMG] allows us to combine P 1 with a prefetcher [ViK] based on the Lempel Ziv data compressor [HoV, Lan, ZiL] to get a prefetcher P that is optimal in the limit against the class of finite state prefetchers. Theorem 2 For every sequence oe n 1 of length n drawn from A, and any s 0, the fault rate of prefetcher P on oe n 1 converges almost surely to Fault F(s) oe n 1 ) as n 1. The expected ....

P. G. Howard and J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression," Information Processing and Management 28 (1992), 749--763, invited paper in special issue on data compression for images and texts.

....information theory [9] applied to predicting [8,12,28] and generalize P 1 to get a universal prefetcher P that is optimal in the limit against a general finite state prefetcher. The resulting optimal prefetcher P is a blend of P 1 and the prefetcher [36] based on the Lempel Ziv data compressor [18,25,37]. We show in Section 6 how to implement the prefetcher in constant expected time per prefetched page, independent of alphabet size ff and cache size k, by use of the optimal dynamic algorithm for generating discrete random variates of Matias, Vitter, and Ni [26] which uses a table lookup method ....

.... the convergence rate cannot be faster than O(1= p n ) 7] The importance of the above theorem lies in its generalization to higher order using techniques from information theory [9] The approach of [12] allows us to combine P 1 with a prefetcher [36] based on the Lempel Ziv data compressor [18,25,37] to get a prefetcher P that is optimal in the limit against the class of finite state prefetchers. Theorem 2 For every sequence oe n 1 of length n drawn from alphabet A, and any s 0, the fault rate of prefetcher P on oe n 1 converges almost surely to Fault F(s) oe n 1 ) as n 1. From ....

P. G. Howard & J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression," Information Processing and Management 28 (1992), 749--763, invited paper in Special Issue on Data Compression for Images and Texts. 20 Krishnan and Vitter

....x 0 = x 1 , x 2 , x c such that for all j 1 substring x j without its last character is equal to some x i , for 0 i j. It encodes the substring x j by the value i, using dlg je bits, followed by the ascii encoding of the last character of x j , using dlg ffe bits. Arithmetic coding [HoV, Lanb, WNC] is a coding technique that achieves a coding length equal to the entropy of the data model. Sequences of probability p are encoded using lg(1=p) bits. Arithmetic coding can be thought of as using fractional bits, as opposed to the suboptimal Huffman coding in which all code lengths must be ....

....E by building a probabilistic model that feeds probability information to an arithmetic coder [BCW, Lana] as explained in the example below. It has been shown that the coding length obtained in this character based approach is at least as good as that obtained using the word based approach [BCW, HoV, Lana]. Hence, the optimality results in [ZiL] hold without change for the character based approach. Example 1 Assume for simplicity that our alphabet is fa; bg. We consider the page access sequence aaaababaabbbabaa . The Ziv Lempel encoder parses this string as (a) aa) ab) aba) abb) b) abaa) ....

[Article contains additional citation context not shown here]

P. G. Howard and J. S. Vitter, "Analysis of Arithmetic Coding for Data Compression," Proceedings of the 1991 IEEE Data Compression Conference (April 1991), invited paper.

No context found.

P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3-12, Snowbird 1991.

No context found.

P. G. Howard and J. Vitter. Analysis of arithmetic coding for data compression. Information Processing and Management, vol 28. no. 6, pages 749-763, 1992.

No context found.

P. G. Howard and J. S. Vitter. Analysis of arithmetic coding for data compression. Information Processing and Management, 28(6):749--763, 1992. Special issue on data compression for images and texts.

No context found.

P. Howard and J. Vitter, Analysis of Arithmetic Coding for Data Compression, Brown University, Department of Computer Science, Proc. Data Compression Conference, 3-12, Snowbird 1991.

No context found.

Howard, P.G. and J.S. Vitter, "Analysis of arithmetic coding for data compression," Proc. Data Compression Conference, pp. 3-12, 1991.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC