scientific article; zbMATH DE number 4021000

From MaRDI portal
Publication:3764034





Image or signal recovery is an ubiquitous problem that is motivated by the desire to get accurate information about the object being acquired despite the limitations of the acquisition system. This book is a collection of thirteen chapters by various authors who have been actively involved with signal recovery problems. The objective of the editor is to introduce the reader to the important results and application areas that have emerged within the last decade. Two fundamental problems that have generated much interest and spurred research during this time are superresolution and phase retrieval, and their application to several areas. The book largely deals with such deterministic solutions to the image recovery problem. Only two chapters deal with statistical techniques. One deals with the Bayesian approach and the other with maximum entropy. The book would perhaps have been more complete if statistical estimation techniques had received more attention. The chapters are well motivated, and well written. Typographical errors were minimal. The chapters are largely self contained, and many have adequate bibliographies. The material draws on diverse areas of mathematics, and some fluency in applied mathematics will be required to fully appreciate the material. The book will be useful to researchers and graduate students in the image processing area, exposing them to some fundamental considerations in image recovery, and the diverse areas in which it finds application.In the following, we review the chapters, generally grouping them where possible by the problems they attempt to solve, rather than in numerical order. The first chapter by C. K. Rushforth introduces the image restoration problem as a solution to the Fredholm equation of the first kind. The ill-posed nature of the problem is discussed, along with regularization techniques to mitigate its severity. The algorithms considered for solution are based on least squares techniques and projection methods, some of them iterative. The chapter also contains a section on Hilbert spaces and linear operators, which is handy in understanding the material to follow. The second chapter by D. C. Youla examines the restoration problem in Hilbert space. The problem is formulated by incorporating apriori information as a set of convex constraints. This leads to iterative algorithms involving a succession of projection operators, which find the solution as a point in the intersection of the convex sets. This method is referred to (in some chapters of the book) as POCS (projection onto convex sets). The reader is led to the algorithm formulation with great care for the mathematical niceties in a systematic fashion. This chapter adds to Youla's well known work on alternating orthogonal projections, which generalized the well known Gerchberg-Papoulis superresolution algorithm. Three chapters in the book address the problem of image recovery from limited data in the computer tomography (CT) context. The approaches and viewpoints taken by the authors is varied, the analytical tools different, but each chapter complements the others quite nicely. A direct application of the POCS method is demonstrated in Chapter 11 by Sezan and Stark. Here the Fourier transform of the signal is only partially specified on the frequency plane. The missing Fourier components must be estimated to reconstruct the object. This task is facilitated by incorporating as much apriori information as possible. This is shown to be easily incorporated in the POCS framework as constraints, and some algorithms are derived. An example of reconstructing a thorax phantom by combining POCS with the direct Fourier method is presented when limited data is available. The authors also briefly cover applications of this technique to areas such as synthetic aperture radar, magnetic resonance imaging, and ultrasound. In Chapter 9, Medoff develops an inner product framework for image reconstruction. It is shown that within this framework, seemingly unrelated algorithms such as convolution backprojection and SIRT (simultaneous iterative reconstruction technique) are unified as yielding least squares solutions under proper choice of norms. The author obtains the reconstructed image in the general case as a minimum norm least squares solution, and provides various iterative techniques to obtain this solution. The problem formulation easily generalizes to the incomplete data case. A fundamental issue here is the possible existence of ghost images, i.e. those images which have no contribution along the data gathering path, but do have contributions elsewhere. This leads to uncertainty in the reconstructed image. Examples are shown where this uncertainty manifests itself as streaks in the reconstruction. Thus additional apriori information such as nonnegativity, known boundary, etc. must be incorporated in the algorithm as a set of constraints. When this is done, the reconstructions do show a great deal of artifact suppression. A Bayesian approach is taken to combat the effects of incomplete data in CT reconstruction in Chapter 3 by Hanson. The chapter contains a detailed discussion of the problems arising from the null space components when a purely deterministic approach is taken. This problem is ameliorated when apriori information is taken into account, and the author provides comparisons among many constrained deterministic algorithms. The Bayesian approach is proposed as technique in which information about the general structure and shape of the object can be incorporated. Within this framework, the author examines two methods, maximum aposteriori (MAP) and fit and iterative reconstruction (FAIR). In the latter technique a parametric model is fitted to the available projection data in order to provide an initial estimate for an iterative scheme (algebraic reconstruction). Both methods have their drawbacks, and a generalized Bayesian scheme is proposed that incorporates the strengths of each method. Numerous simulation examples on phantoms are given to illustrate the theory. Another set of four chapters which can be grouped together consider the problem of retrieving Fourier transform magnitude or phase, given the other. Such problems are encountered in many areas. Some examples: in X- ray crystallography and electron microscopy, only Fourier transform magnitude is available. In order to recover the structure of the object completely, the phase must be retrieved. Conversely, the magnitude retrieval problem arises frequently in radar, sonar, seismology, and ocean acoustics. In Chapter 6 by Hayes, the emphasis is on developing the theory associated with magnitude and phase retrieval of multidimensional sequences. The conditions under which either the Fourier transform magnitude or phase uniquely specifies the other are examined for both the one dimensional and two dimensional cases. It is shown that two dimensional sequences of practical interest are indeed specified by either the magnitude or the phase. Some algorithms are briefly described for implementation. Many examples are presented showing the importance of phase in images. Chapter 7 by Dainty and Fineup focuses on the phases retrieval problem in imaging situations encountered in astronomy. The emphasis is on algorithm design for retrieval. Many algorithms are given, but a Gerchberg-Papoulis type is considered thoroughly, along with several details for implementation. The algorithm fails to converge in some situations. The reasons are identified, and methods to overcome this are presented. In some imaging situations such as speckle interferometry, additional information about the pase at low spatial frequencies is available. Algorithms which incorporate this additional information are given for objet recovery. Many examples are shown with an extensive bibliography. In Chapter 8, Levi and Stark provide algorithms based on generalized projections for magnitude/phase recovery. In the problem of recovery from phase, the constraints are convex, and the POCS method is applicable. Examples are presented to show that some algorithms for recovery from phase may fail to converge while the POCS method has guaranteed (albeit weak) convergence. The convergence rate is shown to improve by optimizing relaxation parameters in the POCS algorithm. For recovery from magnitude, the constraints are non-convex, and a generalization of the POCS algorithm is derived by the authors to deal with this situation. The generalized algorithm may not converge to the true solution, and reasonable conjectures are advanced for this behaviour, though further work is needed to design algorithms that avoid such traps. The role of analyticity in image recovery is presented by Fiddy in Chapter 13. The properties of multidimensional bandlimited functions are described. It is shown that factorizations of these functions provide insight into recovering signals from incompletely specified Fourier data and enable issues such as uniqueness to be answered. The analyticity properties are used to derive algorithms for signal recovery from Fourier magnitude. The methods are computationally intensive, and more research needs to be done to reduce this. In Chapter 4, Mammone considers image restoration using linear programming techniques when the images are sparse, i.e. when the image contains a large number of zero valued samples. Examples arise in astronomy, microscopy, and medical imaging. A linear programming algorithm based on the simplex method is proposed to recover a sparse image from a rank deficient system of equations, when the rank is much smaller than the number of unknowns. An advantage of such a technique is that constraints are part of the algorithm, and thus null space components of the object are not completely lost. In Chapter 5, Zhuang et. al. consider maximum entropy techniques for image recovery. Maximum entropy is a reasonable formulation for signal recovery from incomplete data in the sense that it seeks to maximize the entropy consistent with the given data. This nonlinear method is shown to have a smoothing effect that avoids artifacts usually associated with linear techniques at the expense of considerable computational effort. Three maximum entropy algorithms are discussed. Frieden's technique required the solution of \(m\) nonlinear equations in \(m\) unknowns, which was implemented using the Newton-Raphson technique. When \(m\) is large, this algorithm is not practical. The formulation of Burch et. al. is different, and avoids the large number of Lagrange multipliers introduced in the previous case. This facilitates the use of iterative algorithms, and eases the computational burden. The authors present their own version of the maximum entropy problem in which a system of differential equations is solved to obtain the solution. Instead of optimizing in m dimensional space, the problem reduces in this case to a one dimensional search in m space. Some results are presented, but it is a new algorithm and its performance needs more assessment. In Chapter 10, Kaveh and Soumekh consider computer assisted diffraction tomography. Applications arise when the geometrical optic assumptions of wave propagation in a medium are not valid. Recovery of object information is an inverse problem of the Helmholtz wave equation with complex spatially varying coefficient. This is a difficult proposition and needs simplification before the problem can be tackled and a solution found in a computationally efficient manner. The approximations are obtained using the Born and Rytov transformations. The next step addresses the signal processing issues that make reconstruction feasible. The Fourier slice theory in the context of diffraction tomography is derived, and this forms the basis inversion and reconstruction. Spatial domain and frequency domain reconstruction algorithms are given. Reconstruction from limited angular data is also considered, and points of correspondence with X-ray CT are discussed. Experimental results are shown for ultrasound. The chapter provides a good introduction to the topic and clearly assesses areas for improvement. Finally, in Chapter 12, Saleh departs from the theme of the rest of the book and addresses image synthesis rather than image recovery. Image synthesis seeks to find an input image, which when operated on by an imaging system produces a desired output. As opposed to image recovery, the question of existence is critical. In many cases, this is difficult or impossible to establish. The best one can do is to provide solutions, assuming existence. Three common imaging systems are considered: diffraction limited, diffraction limited with a high contrast film, and a coherent optical system with high contrast detection. Many solution techniques are presented based on linear programming, POCS, and zero crossing theory. Contents: Chapter 1. C. K. Rushforth, Signal Restoration, Functional Analysis, and Fredholm Integral Equations of the First Kind. Chapter 2. D. C. Youla, Mathematical Theory of Image Restoration by the Method of Convex Projections. Chapter 3. Kenneth M. Hanson, Bayesian and Related Methods in Image Reconstruction from Incomplete Data. Chapter 4. Richard Mammone, Image Restoration Using Linear Programming. Chapter 5. Xinhua Zhuang, Einar Østevold and Robert M. Haralick, The Principle of Maximum Entropy in Image Recovery. Chapter 6. Monson H. Hayes, The Unique Reconstruction of Multidimensional Sequences from Fourier Transform Magnitude or Phase. Chapter 7. J. C. Dainty and J. R. Fienup, Phase Retrieval and Image Reconstruction for Astronomy. Chapter 8. Aharon Levi and Henry Stark, Restoration from Phase and Magnitude by Generalized Projections. Chapter 9. Barry P. Medoff, Image Reconstruction from Limited Data: Theory and Applications in Computerized Tomography. Chapter 10. M. Kaveh and M. Soumekh, Computer-Assisted Diffraction Tomography. Chapter 11. M. Ibrahim Sezan and Henry Stark, Applications of Convex Projection Theory to Image Recovery in Tomography and Related Areas. Chapter 12. Bahaa E. A. Saleh, Image Synthesis: Discovery Instead of Recovery. Chapter 13. M. A. Fiddy, The Role of Analyticity in Image Recovery.




Cited in
(only showing first 100 items - show all)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3764034)