A Subpixel Image Restoration Algorithm

Abstract

In statistical image reconstruction, data are often recorded on a regular grid of squares, known as pixels, and the reconstructed image is defined on the same pixel grid. Thus, the reconstruction of a continuous planar image is piecewise constant on pixels, and boundaries in the image consist of horizontal and vertical edges lying between pixels. This approximation to the true boundary can result in a loss of information that may be quite noticeable for small objects, only a few pixels in size. Increasing the resolution of the sensor may not be a practical alternative. If some prior assumptions are made about the true image, however, reconstruction to a greater accuracy than that of the recording sensor's pixel grid is possible. We adopt a Bayesian approach, incorporating prior information about the true image in a stochastic model that attaches higher probability to images with shorter total edge length. In reconstructions, pixels may be of a single color or split between two colors. The model is illustrated using both real and simulated data.     

Evolutionary Simulated Annealing With Application to Image Restoration

Simulated annealing is a randomized algorithm proposed for finding a global optimum in large problems where a target function may have many local extrema. This article considers a modification of the simulated annealing algorithm that turns it into a deterministic technique. Instead of carrying out a stochastic jump based on the annealing update density, the update density is used to select a fixed number of candidate parameter vectors which are all fed to the next iteration of the algorithm. The selection criterion involves not only the update density height, but also information about the origin of the candidate vector. Thus, each iteration produces a cooperative collection of parameter vectors in hope of exploring the parameter space in search of the optimum more thoroughly than the regular annealing. The technique is shown to outperform regular annealing on the problem of restoration of lattice images consisting of simple-shaped objects.     

非线性再生散度随机效应模型的贝叶斯分析

非线性再生散度随机效应模型是一类非常广泛的统计模型，包括了线性随机效应模型、非线性随机效应模型、广义线性随机效应模型和指数族非线性随机效应模型等．本文研究非线性再生散度随机效应模型的贝叶斯分析．通过视随机效应为缺失数据以及应用结合Gibbs抽样技术和Metropolis-Hastings算法（简称MH算法）的混合算法获得了模型参数与随机效应的同时贝叶斯估计．最后，用一个模拟研究和一个实际例子说明上述算法的可行眭．     

Gibbs Measures and Markov Random Fields with Association $$I $$

We introduce the notions of a Gibbs measure with the corresponding potential with association (where is a subset of the set ) of a Markov random field with memory and measure with association . It is proved that these three notions are equivalent.     

Analysis of convergence rates of some Gibbs samplers on continuous state spaces

We use a non-Markovian coupling and small modifications of techniques from the theory of finite Markov chains to analyze some Markov chains on continuous state spaces. The first is a generalization of a sampler introduced by Randall and Winkler, and the second a Gibbs sampler on narrow contingency tables.     

Bayesian Analysis of a Two-State Markov Modulated Poisson Process

Abstract

We postulate observations from a Poisson process whose rate parameter modulates between two values determined by an unobserved Markov chain. The theory switches from continuous to discrete time by considering the intervals between observations as a sequence of dependent random variables. A result from hidden Markov models allows us to sample from the posterior distribution of the model parameters given the observed event times using a Gibbs sampler with only two steps per iteration.     

Computational Bayesian Analysis of Hidden Markov Models

Abstract

Versions of the Gibbs Sampler are derived for the analysis of data from hidden Markov chains and hidden Markov random fields. The principal new development is to use the pseudolikelihood function associated with the underlying Markov process in place of the likelihood, which is intractable in the case of a Markov random field, in the simulation step for the parameters in the Markov process. Theoretical aspects are discussed and a numerical study is reported.     

Reversible algorithm of simulating multivariate densities with multi-hump

To simulate a multivariate density with multi-hump, Markov chain Monte Carlo method encounters the obstacle of escaping from one hump to another, since it usually takes extraordinately long time and then becomes practically impossible to perform. To overcome these difficulties, a reversible scheme to generate a Markov chain, in terms of which the simulated density may be successful in rather general cases of practically avoiding being trapped in local humps, was suggested.     

Efficient Spatial Modeling Using the SPDE Approach With Bivariate Splines

Gaussian fields (GFs) are frequently used in spatial statistics for their versatility. The associated computational cost can be a bottleneck, especially in realistic applications. It has been shown that computational efficiency can be gained by doing the computations using Gaussian Markov random fields (GMRFs) as the GFs can be seen as weak solutions to corresponding stochastic partial differential equations (SPDEs) using piecewise linear finite elements. We introduce a new class of representations of GFs with bivariate splines instead of finite elements. This allows an easier implementation of piecewise polynomial representations of various degrees. It leads to GMRFs that can be inferred efficiently and can be easily extended to nonstationary fields. The solutions approximated with higher order bivariate splines converge faster, hence the computational cost can be alleviated. Numerical simulations using both real and simulated data also demonstrate that our framework increases the flexibility and efficiency. Supplementary materials are available online.     

Conditionally Specified Space-Time Models for Multivariate Processes

This article proposes a class of conditionally specified models for the analysis of multivariate space-time processes. Such models are useful in situations where there is sparse spatial coverage of one of the processes and much more dense coverage of the other process(es). The dependence structure across processes and over space, and time is completely specified through a neighborhood structure. These models are applicable to both point and block sources; for example, multiple pollutant monitors (point sources) or several county-level exposures (block sources). We introduce several computational tricks that are integral for model fitting, give some simple sufficient and necessary conditions for the space-time covariance matrix to be positive definite, and implement a Gibbs sampler, using Hybrid MC steps, to sample from the posterior distribution of the parameters. Model fit is assessed via the DIC. Predictive accuracy, over both time and space, is assessed both relatively and absolutely via mean squared prediction error and coverage probabilities. As an illustration of these models, we fit them to particulate matter and ozone data collected in the Los Angeles, CA, area in 1995 over a three-month period. In these data, the spatial coverage of particulate matter was sparse relative to that of ozone.     

Convergence rates of the blocked Gibbs sampler with random scan in the Wasserstein metric

ABSTRACT

This paper establishes explicit estimates of convergence rates for the blocked Gibbs sampler with random scan under the Dobrushin conditions. The estimates of convergence in the Wasserstein metric are obtained by taking purely analytic approaches.     

Nonlinear PDE Model for Image Restoration Using Second-Order Hyperbolic Equations

In this article we consider a novel nonlinear PDE-based image denoising technique. The proposed restoration model uses second-order hyperbolic diffusion equations. It represents an improved nonlinear version of a linear hyperbolic PDE model developed recently by the author, providing more effective noise removal results while preserving the edges and other image features. A rigorous mathematical investigation is performed on this new differential model and its well-posedness is treated. Next, a consistent finite-difference numerical approximation scheme is proposed for this nonlinear diffusion-based approach. Our successful image denoising experiments and method comparisons are also described.     

On Dynkin's Markov property of random fields associated with symmetric processes

Let p(t, x, y) be a symmetric transition density with respect to a σ-finite measure m on (E, $\text{E}$), g(x,y)=∫p(t,x,y)dt, and $\text{M}\text{={σ-finite measures μ?0:∫g(x,y)μ(}\text{d}\text{x)μ(}\text{d}\text{y)<∞}}$. There exists a Gaussian random field $\text{Φ={?}{}_{\mathrm{\mu }}\text{:μ?}\text{M}\text{}}$ with mean 0 and covariance $\text{E}\text{?}{}_{\mathrm{\mu }}\text{?}{}_{\mathrm{\nu }}\text{=∫g(x,y)μ(}\text{d}\text{x)ν(}\text{d}\text{y)}$. Letting $\text{F}\text{(B)=σ{?}{}_{\mathrm{\mu }}\text{:μ(B}{}^{\mathrm{c}}\text{)=0}}$ we consider necessary and sufficient conditions for the Markov property (MP) on sets B, C: $\text{F}$(B), $\text{F}$(C) c.i. given $\text{F}$(B ∩ C). Of crucial importance is the following, proved by Dynkin: $\text{E}\text{{?}{}_{\mathrm{\mu }}\text{∣}\text{F}\text{(B)}=?}{}_{\mathrm{\mu B}}$, where μ_B is the hitting distribution of the process corresponding to p, m with initial law μ. Another important fact is that ?_μ=?_ν iff μ, ν have the same potential. Putting these together with an additional transience assumption, we present a potential theoretic proof of the following necessary and sufficient condition for (MP) on sets B, C: For every x?E, T_B∩C=T_B+T_C∮ θ_TB=T_C+T_B∮θ_TC a.s. P^x where, for D ? $\text{E}$, T_D is the hitting time of D for the process associated with p, m. This implies a necessary condition proved by Dynkin in a recent preprint for the case where B∪C=E and B, C are finely closed.     

Image segmentation by polygonal Markov Fields

This paper advocates the use of multi-coloured polygonal Markov fields for model-based image segmentation. The formal construction of consistent multi-coloured polygonal Markov fields by Arak–Clifford–Surgailis and its dynamic representation are specialised and adapted to our context. We then formulate image segmentation as a statistical estimation problem for a Gibbsian modification of an underlying polygonal Markov field, and discuss the choice of Hamiltonian. Monte Carlo techniques, including novel Gibbs updates for the Arak model, to estimate the model parameters and find an optimal partition of the image are developed. The approach is applied to image data, the first published application of polygonal Markov fields to segmentation problems in the mathematical literature. Work carried out under project PNA4.3 ‘Stochastic Geometry’. This research was supported by the EC 6th Framework Programme Priority 2 Information Society Technology Network of Excellence MUSCLE (Multimedia Understanding through Semantics, Computation and Learning; FP6-507752), and partially by the Foundation for Polish Science (FNP) and by the Polish Minister of Scientific Research and Information Technology, grant 1 P03A 018 28 (2005-2007) [the third author].     

Convergence rates of symmetric scan Gibbs sampler

We consider the symmetric scan Gibbs sampler, and give some explicit estimates of convergence rates on the Wasserstein distance for this Markov chain Monte Carlo under the Dobrushin uniqueness condition.     

Geometric ergodicity of the Gibbs sampler for the Poisson change-point model

Poisson change-point models have been widely used for modelling inhomogeneous time-series of count data. There are a number of methods available for estimating the parameters in these models using iterative techniques such as MCMC. Many of these techniques share the common problem that there does not seem to be a definitive way of knowing the number of iterations required to obtain sufficient convergence. In this paper, we show that the Gibbs sampler of the Poisson change-point model is geometrically ergodic. Establishing geometric ergodicity is crucial from a practical point of view as it implies the existence of a Markov chain central limit theorem, which can be used to obtain standard error estimates. We prove that the transition kernel is a trace-class operator, which implies geometric ergodicity of the sampler. We then provide a useful application of the sampler to a model for the quarterly driver fatality counts for the state of Victoria, Australia.     

Reconstructibility of a general DNA evolution model

In this paper, we analyze the tree reconstruction problem, to identify whether there is non-vanishing information of the root, as the level of the tree goes to infinity. Although it has been studied in numerous contexts, the existing literature with rigorous reconstruction thresholds established are very limited, and it becomes extremely challenging when the model under investigation has 4 states, one of whose interpretations is the four main bases found in Deoxyribonucleic acid (DNA) and Ribonucleic acid (RNA): guanine [G], cytosine [C], adenine [A], and thymine [T]. In this paper, we study a general DNA evolution model, which distinguishes between transitions and transversions, and allow transversions to occur at the same rate but that rate can be different from the rates for transitions. The sufficient condition for reconstruction is rigorously established.     

Directed Markov Point Processes as Limits of Partially Ordered Markov Models

In this paper, we consider spatial point processes and investigate members of a subclass of the Markov point processes, termed the directed Markov point processes (DMPPs), whose joint distribution can be written in closed form and, as a consequence, its parameters can be estimated directly. Furthermore, we show how the DMPPs can be simulated rapidly using a one-pass algorithm. A subclass of Markov random fields on a finite lattice, called partially ordered Markov models (POMMs), has analogous structure to that of DMPPs. In this paper, we show that DMPPs are the limits of auto-Poisson and auto-logistic POMMs. These and other results reveal a close link between inference and simulation for DMPPs and POMMs.     

Tracking the output of a poorly known markov process

Tracking the output of an unknown Markov process with unknown generator and unknown output function is considered. It is assumed the unknown quantities have a known prior probability distribution. It is shown that the optimal control is a linear feedback in the tracking error plus the conditional expectation of a quantity involving the unknown generator and output function of the Markov process. The results also have application to Bayesian identification of hidden Markov models     

Estimating Mixture of Dirichlet Process Models

Abstract

Current Gibbs sampling schemes in mixture of Dirichlet process (MDP) models are restricted to using “conjugate” base measures that allow analytic evaluation of the transition probabilities when resampling configurations, or alternatively need to rely on approximate numeric evaluations of some transition probabilities. Implementation of Gibbs sampling in more general MDP models is an open and important problem because most applications call for the use of nonconjugate base measures. In this article we propose a conceptual framework for computational strategies. This framework provides a perspective on current methods, facilitates comparisons between them, and leads to several new methods that expand the scope of MDP models to nonconjugate situations. We discuss one in detail. The basic strategy is based on expanding the parameter vector, and is applicable for MDP models with arbitrary base measure and likelihood. Strategies are also presented for the important class of normal-normal MDP models and for problems with fixed or few hyperparameters. The proposed algorithms are easily implemented and illustrated with an application.