entropy-constrained scalar quantizers

Top PDF entropy-constrained scalar quantizers:

Error bounds for high-resolution quantization with Rényi - α - entropy constraints

Error bounds for high-resolution quantization with Rényi - α - entropy constraints

best approximation of a given probability measure by another probability mea- sure with reduced complexity. Complexity constraints used so far are restricted memory size resp. restricted Shannon entropy of the approximation. The ap- proximating probability is always induced by a quantizer, which decomposes the space into codecells. Every point of a codecell will be mapped by the quan- tizer to a codepoint which is unique for each codecell. The set of all codepoints is called codebook. The mathematical aspects of quantization in finite dimen- sion with restricted memory size have been investigated by Graf, Luschgy et.al. [8, 13, 14], Gruber [18], Dereich et.al. [10] and Fort, Pag` es et.al. [9, 11]. A thor- ough mathematical treatment of (Shannon-)entropy-constrained quantization also emerged in the last few years and has been carried out by Gray, Gy¨ orgy, Linder and Li (see e.g. [16, 19, 20, 21] and the references therein). Sullivan [28] developed an algorithm for designing entropy-constrained scalar quantizers for the exponential and Laplace distribution. The question if optimal entropy- constrained quantizers induce a finite or infinite number of codecells has been investigated by Gy¨ orgy, Linder, Chou and Betts [22]. Recently, quantization has also been studied with combined entropy and memory size constraints (cf. [15]). Apart from studying high-resolution asymptotics also the asymptotic be- havior of the optimal quantization error under small bounds on the complexity constraint has been investigated by several authors (cf. [24, 25, 26]).
Show more

16 Read more

Entropy Density and Mismatch in High-Rate Scalar Quantization with Rényi Entropy Constraint

Entropy Density and Mismatch in High-Rate Scalar Quantization with Rényi Entropy Constraint

In this paper we extend some of the more refined results of fixed and variable-rate asymptotic quantization theory in the framework of quantization with R´enyi entropy constraint of order α ∈ (0, 1). The concept of a quantizer point density is a problematic one for (R´enyi) entropy-constrained quantization since (near) optimal quantizers can have an arbitrarily large number of levels in any bounded region. Instead, we investi- gate the R´enyi entropy contribution of a given interval to the overall rate. One of our main results, Theorem 2, shows that for a large class of source densities and an asymptotically opti- mal sequence of quantizers, this contribution can be quantified by the so called entropy density of the sequence. A dual of this result, Corollary 1, quantifies the distortion contribution of a given region to the overall distortion in terms of the so-called distortion density. Interestingly, it turns out that the entropy and distortion densities are equal in the cases we investigate (Remark 5). Our other main contribution, Theorem 3, is a mismatch formula for a sequence of asymptotically optimal R´enyi entropy constrained scalar quantizers. From our density and mismatch results we can recover the known results for the traditional rate definitions by formally setting α = 0 or α = 1. The rest of the paper is organized as follows. In the next section we formulate the quantization problem and give a somewhat informal overview of our results in the context of prior work. In Section III the entropy and distortion density results are presented and proved. The mismatch problem is considered in Section IV. Concluding remarks are given in Section V.
Show more

12 Read more

Entropy-constrained scalar quantization with a lossy-compressed bit

Entropy-constrained scalar quantization with a lossy-compressed bit

The Lb-ECSQ method resembles works in the field of source-channel coding—namely, channel-optimized quantization [10]. Interestingly, when the output of a scalar quantizer is coded and transmitted via a very noisy channel, quantizers with a small number of levels (higher distortion) may yield better performance than those with a larger number of levels (lower distortion) [11]. Several works have addressed the design of scalar quantizers for noisy channels (e.g., [10–12]). All these works present conditions and algorithms to optimize the scalar quantizer given that it is followed by a noisy channel. This is similar to the Lb-ECSQ setup, where the lossy binary encoder behaves like a “noisy channel”, with an important and critical difference: in our problem, the distortion introduced by the lossy encoder (the“error probability” of the channel) is a parameter to be optimized, and acts as an additional degree of freedom. Note also that we solely consider the problem of source coding of a continuous source; encoded symbols are transmitted errorless to the receiver that aims at reconstructing the source.
Show more

12 Read more

Low-resolution scalar quantization for Gaussian sources and squared error

Low-resolution scalar quantization for Gaussian sources and squared error

To determine the low-resolution performance, we analyze the oper- ational rate-distortion function, R(D), of entropy-constrained scalar quantization in the low-rate region. We focus on squared-error distortion and stationary memoryless sources with absolutely continuous distri- butions, which are completely characterized by the probability density function (pdf) of an individual random variable. Accordingly, R(D) is defined to be the least output entropy of any scalar quantizer with mean- squared error D or less. As it determines the optimal rate-distortion per- formanceof this kind of quantization, it is important to understand how R(D) depends on the source pdf and how it compares to the Shannon rate-distortion function. For example, the performance of conventional transform coding, which consists of an orthogonal transform followed by a scalar quantizer for each component of the transformed source vector, depends critically on the allocation of rate to component scalar quantizers, and the optimal rate allocation is determined by the operational rate-distortion functions of the components [3, p. 227].
Show more

9 Read more

Distributed transform coding via source-splitting

Distributed transform coding via source-splitting

Transform coding (TC) is one of the best known practical methods for quantizing high-dimensional vectors. In this article, a practical approach to distributed TC of jointly Gaussian vectors is presented. This approach, referred to as source-split distributed transform coding (SP-DTC), can be used to easily implement two terminal transform codes for any given rate-pair. The main idea is to apply source-splitting using orthogonal-transforms, so that only Wyner-Ziv (WZ) quantizers are required for compression of transform coefficients. This approach however requires optimizing the bit allocation among dependent sets of WZ quantizers. In order to solve this problem, a low-complexity tree- search algorithm based on analytical models for transform coefficient quantization is developed. A rate-distortion (RD) analysis of SP-DTCs for jointly Gaussian sources is presented, which indicates that these codes can significantly outperform the practical alternative of independent TC of each source, whenever there is a strong correlation between the sources. For practical implementation of SP-DTCs, the idea of using conditional entropy constrained (CEC) quantizers followed by Slepian-Wolf coding is explored. Experimental results obtained with SP-DTC designs based on both CEC scalar quantizers and CEC trellis-coded quantizers demonstrate that actual implementations of SP-DTCs can achieve RD performance close to the analytically predicted limits.
Show more

15 Read more

Exergy concept and exergetic analysis: review

Exergy concept and exergetic analysis: review

entropy and exergy concepts come from thermodynamics and are applicable to all fields of science and engineering. Therefore, this article intends to provide background for better understanding of these concepts and their differences among various classes of life support systems a comprehensive and critical view on is presented. It also covers the basic principles, general and implications in various fields and its characteristic has been discussed and different forms of exergy have been derived. The exergy of an energy form or a substance is a measure of its usefulness or quality or potential to cause change. Exergy is naturally related to the concept of quality of energy. Therefore, exergy analysis has been widely applied in parallel with energy analysis in order to find the most rational use of energy. Also a brief comparison Finally, conclusions regarding the usability of the exergy method as a tool to promote a more efficient use of available energy sources are also derived.
Show more

8 Read more

Construction of Global Weak Entropy Solution of Initial Boundary Value Problem for Scalar Conservation Laws with Weak Discontinuous Flux

Construction of Global Weak Entropy Solution of Initial Boundary Value Problem for Scalar Conservation Laws with Weak Discontinuous Flux

The present paper is organized as follows. In Section 2, we introduce the definition of weak entropy solution and the boundary entropy condition for the initial-boundary value problem (1), and give a lemma to be used to construct the piecewise smooth solution of (1). In Section 3, basing on the analysis method in [27], we use the lemma on piecewise smooth solution given in Section 2 to construct the global weak entropy solution of the initial-boundary value problem (1) with two pieces of constant initial data and constant boundary data under the condition that the flux function has a finite number of weak discontinuous points, and state the geometric structure and the behavior of boundary for the weak entropy solution.
Show more

18 Read more

Turbulent combustion simulation in realistic gas-turbine combustors

Turbulent combustion simulation in realistic gas-turbine combustors

Thus, Pierce and Moin [88-90] recently proposed a partially premixed combustion model based on the combined mechanisms described above to predict local extinction and re-ignition. Their model was based on a flamelet/progress variable approach (FPVA) so that the scalar properties are now a function of both mixture fraction and progress variable, which describes the extent of reaction in the local mixture. The effect of strain rate as included in the 2 nd mechanism was taken into account through implicitly varying the amount of progression made in chemical reactions. In other words, the progress variable varies with scalar dissipation/strain rate. Their methods is similar to that used by Janicka and Kollmann [91] who solved two transport equations for mixture fraction and a reactive scalar, and closed the chemical reaction term using the transported PDF method. A similar idea was also used by Bruel et al [92] using a presumed-shape PDF method.
Show more

202 Read more

Multipath estimation based on modified ε-constrained rank-based differential evolution with minimum error entropy

Multipath estimation based on modified ε-constrained rank-based differential evolution with minimum error entropy

The PDF-shaping control methodology has been developed and applied to control systems with non-Gaussian noise and nonlinear dynamics [19, 20]. An alternative measure for general non-Gaussian systems is the entropy, which is a scalar quantity in information theory that quantifies the average uncertainty involved in a random variable [21]. Entropy can be used to depict the higher-order statistics of a distribution since it is formulated on the PDF. The use of entropy is not limited to Gaussian assumption. Thus, the so-called minimum error entropy (MEE) criterion has been employed in many stochastic distribution control problems [22-24].
Show more

15 Read more

Mixing of scalar glueballs and flavour-singlet scalar mesons

Mixing of scalar glueballs and flavour-singlet scalar mesons

We explored this mixing in the quenched approximation 共 see Tables I and II 兲 using SW-clover valence quarks of two different masses 关 8 兴 . The zero-momentum glueball operators were measured at every time slice in the usual way 关 10 兴 and the disconnected quark loops are measured as described in the appendix, namely with sufficient stochastic samples that no significant error arises from the stochastic algorithm. The connected quark correlators were taken from previous mea- surements 关 8 兴 . Since the scalar meson or glueball has vacuum quantum numbers, we subtract the vacuum contribu- tion in the other types of correlation we measure. Our results for all of these types of correlation are illustrated in Fig. 2 for the case of one choice of glueball operator and one 共 local 兲 mesonic operator at our lighter quark mass.
Show more

11 Read more

Monotonic Estimation for Probability Distribution and Multivariate Risk Scales by Constrained Minimum Generalized Cross Entropy

Monotonic Estimation for Probability Distribution and Multivariate Risk Scales by Constrained Minimum Generalized Cross Entropy

of (3.11) subject to (3.12) are the solution to the isotonic regression problem where weights 𝑤 𝑖 are equal to 𝑛. The algorithm PAV repeatedly searches both backward and forward for violators and takes average whenever a violator is found. In contrast, Algorithm 4.1 determines explicitly the groups of consecutive indexes by a forward search for partition integers. Average is then to be taken over each of these groups. For Algorithm 4.2, the constrained optimization is transformed into a non-constrained mathematical programming, through a re-

10 Read more

Model for the collection and analysis of data from teachers and students, supported by Academic Analytics

Model for the collection and analysis of data from teachers and students, supported by Academic Analytics

Scalar Average Teacher Rating. Scalar Number of active semesters. Scalar Number of courses taught. Scalar Average grades obtained by students. Scalar Students' approval rate. Chart [r]

21 Read more

Using Geometrical Properties for Fast Indexation of Gaussian Vector Quantizers

Using Geometrical Properties for Fast Indexation of Gaussian Vector Quantizers

For this reason, most of heuristic algorithms we can find in literature first perform a deterministic search in a set of admissible configurations and then, in order not to termi- nate in a local minimum of a cost function, adopt a random- ized approach (e.g., randomly generating the next configu- ration and allowing within reasonable limits the configura- tions of higher cost than the present). Among the first pa- pers that assessed an index assignment problem for vector quantization by an heuristic approach one can find those of De Marca and Jayant [7], and of Chen et al. [19]. Farvardin [20] employed to the problem a simulated annealing algo- rithm. Zeger and Gersho [12] proposed a binary switching method, where pairs of codevectors change index in an iter- ative fashion, determined by a cost function. Potter and Chi- ang [18] presented a paper using minimax criterion based on hypercube that improves the worst case performance, im- portant for the image perception. Knagenhjelm and Agrell employed the Hadamard transform to derive first an objec- tive measure on success of index assignment [10] and then, to design efficient index assignment algorithms [11]. Similar theory was applied by Hagen and Hedelin [16] for designing vectors quantizers with good index assignments.
Show more

11 Read more

Towards optimal symbolization for time series comparisons

Towards optimal symbolization for time series comparisons

Symbolic Aggregation Approximation is a time series rep- resentation that actually supports an arbitrary underlying quan- tizer [5, pg. 59] (as denoted by Q in equation 1). However, in their research SAX’s authors have chosen to use a quantizer based on MOE, but with the added assumption of a normal distribution [1]. It is this quantizer variant used within SAX that we consider, henceforth denoted as qSAX. Note we only address the performance of this specific part of the SAX representation, and not the other aspects such as temporal quantization or the efficient indexing of the symbolic repre- sentation addressed in subsequent publications (e.g. extended SAX [4], iSAX [13] or iSAX 2.0 [5], which continue to use qSAX as the underlying quantizer). Importantly, all of the quantizers evaluated in this work could also be used in any of the overall SAX frameworks, allowing any performance improvements we report to also benefit these more involved approaches to working with time series data.
Show more

9 Read more

Theory and simulation of passive scalar mixing in the presence of a mean scalar gradient

Theory and simulation of passive scalar mixing in the presence of a mean scalar gradient

plied in an Eulerian frame of reference, the resulting equations are not consistent with the Kolmogorov form for the energy spectrum in the inertial range. However, there are several Lagrangian reformulations of the DIA without this property, one example being the sparse direct-interaction perturbation (SDIP), first introduced by Kida and Goto [24] with the name Lagrangian direct-interaction approximation. It is a renormalized closure theory for second-order turbulent statistics that applies a similar procedure to Kraichnan’s direct-interaction approximation (DIA) [25] in a Lagrangian framework. The SDIP is simpler than the Lagrangian history DIA of Kraichnan [26], and yields the same integro-differential equations as the Lagrangian renormalized approximation (LRA) of Kaneda [21]. The SDIP has been used to cal- culate the energy spectrum [24], and the scalar spectrum [14]. Goto and Kida [15] applied the SDIP to a simpler dynamical model to better understand the basis of the approximation. In light of the importance of sparse coupling in the approxi- mation, the name sparse direct-interaction perturbation was then chosen in place of Lagrangian direct-interaction approximation. Here, we will use the SDIP to calculate the velocity-scalar cospectrum. One advantage that the SDIP has over simulation and experiment is the relative ease with which the cospectrum can be calculated for a range of different Schmidt numbers.
Show more

144 Read more

Content Based Image Retrieval in Art Collection Using Edbtc

Content Based Image Retrieval in Art Collection Using Edbtc

Content based image retrieval(CBIR), also known as query by image content(QBIC).This paper presents a new approach to derive the image feature descriptor from the Error-diffusion based block truncation coding (EDBTC) compressed data stream. The image feature descriptor is simply constructed from two EDBTC representative color quantizers and its corresponding bitmap image. The color histogram feature (CHF) derived from two color quantizers represents the color distribution and image contrast, while the bit pattern histogram feature (BHF) constructed from the bitmap image characterizes the image edges and textural information.The similarity between two images can be easily measured from their CHF and BHF values using a specific distance metric computation. This CBIR technique can be implemented on Art Collection. Here all painting of an artist is stored as a database one painting is selected as query image. Same query image is given for recognition, it will recognize the artist of that painting. Also it displays other paintings of that artist.
Show more

7 Read more

A Scalar Compromise Equilibrium for N Person Prescriptive Games

A Scalar Compromise Equilibrium for N Person Prescriptive Games

Example 4.2. Consider now the discoordination game [16] given by Figure 3. Players 1 and 2 approach each other. Player 1’s incentive is to veer right (R) or left (L) in the opposite direction from Player 2’s move. Howev- er, Player 2’s incentive is to encounter Player 1. There is no pure NE or mutual-max outcome for this game. The maximin outcome is (R, R). The CE scalar values are shown in Figure 4, with the unique CE being the Pareto optimum (R, R). The maximin outcome is the same as the CE.

6 Read more

Chapter07_1 (2).ppt

Chapter07_1 (2).ppt

The second law of thermodynamics leads to the definition of a new property called entropy, a quantitative measure of microscopic disorder for a system. Entropy is a measure of energy that is no longer available to perform useful work within the current environment. To obtain the working definition of entropy and, thus, the second law, let's derive the Clausius inequality.

51 Read more

Wavelet Scalar Quantization

Wavelet Scalar Quantization

wavelet scalar quantization (WSQ) is a lossy wavelet compression. We are carefully designed, it can satisfy the criteria above and results are efficient compression, where important small details are preserved, or at least identifiable. Fingerprint images exhibit characteristic high energy in certain high frequency bands resulting from the ridge valley pattern and other structures. These minutiae characteristics are local discontinuities in the fingerprint pattern which represent terminations and bifurcations. A ridge termination is defined as the point at which a ridge ends abruptly. A ridge bifurcation is defined as the point at which a ridge forks or diverges into branch ridges. The ridge structure in fingerprint images is not always well defined, and therefore, an enhancement is required before compression [5]. In our work enhancement is carried out using local histogram equalization and wiener filtering and image binarization. To account for this property, the wavelet scalar quantization (WSQ) standard for lossy fingerprint compression uses a specific wavelet packet sub band structure, which emphasis the important high frequency bands [1]. Filter choice is classical pyramidal coding scheme specifically tuned for fingerprint compression. It is identified that biorthogonal filters are being superior to orthogonal filters and also they optimize filters for fingerprint compression. In this work a
Show more

6 Read more

A Study of trellis coded quantization for image compression

A Study of trellis coded quantization for image compression

and is the 8-state trellis with a quantifiable quantizers rate of quantization this thesis is to investigate conjunction with image transforms today; the discrete Trellis tech evolved as[r]

96 Read more

Show all 5687 documents...