Font Size: a A A

Evolutionary Optimization And Ts Applications In Image Restoration

Posted on:2003-10-17Degree:DoctorType:Dissertation
Country:ChinaCandidate:S Y CengFull Text:PDF
GTID:1118360182993871Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Evolutionary Computation is a stochastic optimization strategy andtechnique, which gets inspirations from natural evolution. Because of itsproperties of robustness, universality, self-organizing, self-adaptation andself-learning, the theoretical research is involved increasingly deep, and theapplication becomes increasingly large. This thesis introduces partiallyordered relation into evolutionary algorithm, which clues our study in researchand application of evolutionary algorithm.In chapter 1, we introduce the origination and development ofevolutionary computation, describe the general process when using it to solvea problem, and summarize its main advantages and application prospect.In chapter 2, partially ordered relation is introduced into evolutionaryalgorithm, where we generalize the basic structure for evolutionary algorithm,which enlarges the conception of evolutionary algorithm and guides widerregion of its application.In chapter 3, we design an evolutionary algorithm of contracting searchspace based on partially ordered relation for constrained optimizationproblems. Theoretical study guarantees the convergence of the new algorithm,and the numerical results demonstrate that the algorithm is superior to othermethods in terms of solution quality, robustness and convergent rate.Especially, for some classical engineering optimal problems, the first digitalsof the values of the resulting solution from our algorithm are better than othernewly designed algorithms, such as IDCNLP, SA, MVEP, MIHDE, and so on.In chapter 4, we design an orthogonal multi-objective evolutionaryalgorithm (OMOEA) for multi-objective optimization problems (MOPs) withconstraints, which uses a determinate search without randomness. We takefour benchmark problems designed by Deb and an engineering problem(WATER) with constraints provided by Ray et al. to test our algorithm. Thenumerical experiments show that our algorithm is superior to other MOGASand MOEAs, such as FFGA, NSGAII, SPEA, SPEA2, and so on, in terms ofthe precision, the quantity and the distribution uniformity of solutions and theconvergence rate of algorithm. Especially, for the engineering problemWATER, it finds the Pareto-optimal set which was previously unknown.In chapter 5, an evolutionary algorithm is designed to apply toregularization method so as to solve ill-posed problems of image restoration.Theoretical analysis and numerical results show that the regularized imageresulted from the evolutionary algorithm is better than that of other methodscurrently known. At the same time, we find that the optimal regularizationoperator should usually be lowstop and highpass, and minimizing the normof the residue of restoration error yield an approximate optimalregularization parameter.
Keywords/Search Tags:evolutionary computation, partially ordered relation, function optimization, multi-objective optimization, image restoration
PDF Full Text Request
Related items