Minimum cross-entropy methods in image superresolution

In medical imaging applications, the expectation maximization (EM) algorithm is a popular technique for obtaining the maximum likelihood estimate (MLE) of the solution to the inverse imaging problem. The Richardson/Lucy (RL) method, derived under different assumptions, is identical to this particular EM algorithm. The RL method is commonly used by astronomers in image deconvolution problems from astronomical data. A closely related algorithm, which we shall refer to as the Poisson MLE, was proposed recently in the context of image superresolution. These algorithms can be grouped under minimum Kullback-Leibler distance methods (minimum cross-entropy methods) as opposed to the standard least-squares methods. The purpose of this paper is twofold. In the first part we explore a common underlying conceptual similarity in the algorithms, even though they were derived under varying assumptions. In the second part, we empirically evaluate the performance of this class of algorithms via experiments on simulated objects, for the image superresolution problem. One set of experiments examines the data consistency performance of the algorithms. A second set of experiments evaluates the performance on the addition of simple constraints on the estimate.

[1]  P. L. Combettes The foundations of set theoretic estimation , 1993 .

[2]  L. Lucy An iterative technique for the rectification of observed distributions , 1974 .

[3]  E. Kreyszig Introductory Functional Analysis With Applications , 1978 .

[4]  William H. Richardson,et al.  Bayesian-Based Iterative Method of Image Restoration , 1972 .

[5]  Edward S. Meinel,et al.  Origins of linear and nonlinear recursive restoration algorithms , 1986 .

[6]  Charles L. Byrne,et al.  Iterative image reconstruction algorithms based on cross-entropy minimization , 1992, Optics & Photonics.

[7]  T. Holmes Expectation-maximization restoration of band-limited, truncated point-process intensities with application in microscopy , 1989 .

[8]  Timothy J. Holmes Maximum-likelihood image restoration adapted for noncoherent optical imaging , 1988 .

[9]  B R Frieden Image recovery by minimum discrimination from a template. , 1989, Applied optics.

[10]  L. Bregman The relaxation method of finding the common point of convex sets and its application to the solution of problems in convex programming , 1967 .

[11]  Y. Censor,et al.  On Block-Iterative Entropy Maximization , 1987 .

[12]  L. Shepp,et al.  Maximum Likelihood Reconstruction for Emission Tomography , 1983, IEEE Transactions on Medical Imaging.

[13]  Robert M. Gray,et al.  Minimum Cross-Entropy Pattern Classification and Cluster Analysis , 1982, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[14]  D. Luenberger Optimization by Vector Space Methods , 1968 .

[15]  Charles L. Byrne,et al.  General entropy criteria for inverse problems, with applications to data compression, pattern classification, and cluster analysis , 1990, IEEE Trans. Inf. Theory.

[16]  K. Lange,et al.  A Theoretical Study of Some Maximum Likelihood Algorithms for Emission and Transmission Tomography , 1987, IEEE Transactions on Medical Imaging.

[17]  B. Hunt,et al.  Analysis of the limit to superresolution in incoherent imaging , 1993 .

[18]  J. J. Burke,et al.  Estimating objects from their blurred and grainy images , 1975 .

[19]  Bobby R. Hunt,et al.  Description of a Poisson Imagery Super Resolution Algorithm , 1992 .

[20]  R. Mersereau,et al.  Iterative methods for image deblurring , 1990 .