AuthorCipperly, George Edward.
Committee ChairFrieden, B. Roy
MetadataShow full item record
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractSeveral aspects of extracting scene object information directly from the associated autocorrelation (or spectrum modulus) data arrays are investigated. Emphasis is on the particular scenario in which the scene can be modelled by a small set of dispersed objects with associated position, size, shape, and brightness parameters. These parameters may completely define a scene, they may contain the information of interest in a more complex scene, or they may merely constitute a reasonable first approximation to an arbitrary scene. A typical two step approach to estimating such parameters is to first use phase retrieval/image reconstruction techniques to estimate an associated image, and then apply pattern recognition techniques to extract the important information from it. The work described here focusses on eliminating the image recovery step and estimating parameter values directly from the autocorrelation. This task naturally separates into several distinct sub-problems, the first of which is extracting the significant features from the autocorrelation. This is a pattern recognition problem with special consideration to the unique features of autocorrelations. From these feature positions, the number of objects in the scene and their relative positions are next deduced, and finally, the individual object sizes, shapes and brightnesses are extracted. Optional further analysis is described in which the object parameter estimates are further refined by seeking a Maximum Likelihood estimate with regard to the data array. Alternatively, the initial estimates could be used to generate a trial image for an iterative phase retrieval procedure to reconstruct the full scene. Since the trial estimate already contains the major features of the scene, convergence to the correct solution should be both faster and better assured. The phase retrieval problem has been well studied and is not investigated here. For each of these sub-problems, the logical or mathematical development of the solution is presented, implementing computer algorithms are described, and theoretical and practical limitations are discussed.
Degree ProgramOptical Sciences