Dissertation > Excellent graduate degree dissertation topics show

Genetic Algorithm and Its Application in Image Segmentation

Author: PengXuan
Tutor: LiuBo
School: Jilin University
Course: Basic mathematics
Keywords: Evolutionary Computation Genetic Algorithms Image Segmentation
CLC: TP391.41
Type: Master's thesis
Year: 2006
Downloads: 386
Quote: 2
Read: Download Dissertation

Abstract

This article has carried on not only the detailed elaboration to the geneticalgorithms theory, but also positive and effective search to the application inimage segmentation.The genetic algorithm is a kind of stochastic search algorithm, whichconsults model biosphere natural selection and nature heredity mechanism, canobtain related knowledge of the search space in the automatic search processand auto-adapted control the search process, then achieves the optimal solutionor the accurate optimal solution. It has some good characteristic such assimpleness, generality, robustness, broad scope is suitable to paralleldistribution processing. The genetic algorithms may be used in the imagesegmentation, and obtained the good division effect. The use of geneticalgorithms in the image processing, on the one hand, has provided one kind ofnew mentality for the imagery processing, on the other hand also expanded thegenetic algorithms application domain.First chapter mainly narrated the simulation evolution computationtechnology. The simulation evolution computation technology (SEC) simulatesthe nature biology evolution process and the mechanism, is a kind ofauto-adapted artificial intelligence optimizes and search technology.This core thought of the technology is such elementary knowledge: Thebiological evolution process (from preliminaryility to senior, from simpleness tocomplexness,) is natural, parallel steady optimized process itself. This goal ofthe optimized process is to enable the life body to achieve the best structure andeffect for environment, and the biological population evolutes through thetheory of "superior win and the inferior wash out" and the principle ofhereditary and mutation.SEC is one kind of more macroscopic significance biological modellingoptimization algorithm, which imitates all lives and intelligent production andevolution process. It not only simulates Darwin’s evolution principle of"superior win and the inferior wash out, survival of the fittest", but also seeks abetter structure through simulating Mendel’s hereditary and mutation theory .To each body A, SEC should assign its corresponding sufficiencyaccording to the certain rule. In general, the rule of fitness decision should makeindividual fitness relate with the goal function value f of individual phenotypeX. The fitness is more, the objective function approaches the global maximumvalue. Otherwise, the fitness is smaller. The rule definition is as follow.Supposes R+0 is a real set, a mapping F : ? →R0+ is called the fitnessfunction of question (1.1), if F and f have the same overall situation maximumvalue, and satisfies: ( 1)(2)0(1)(2)f X ≥ fX≥?FX≥FX.In general,the simulation evolution algorithm approach by the next foursteps composedStep 1 (initialization) determined colony scale N and the terminationcriterion (for example establishment the biggest evolution algebra or expectapproximate solution precision which achieved);Stochastically produces Nindividual to take the initial colony x(0)→;Set at the evolution algebra countert →0.Step 2 (individual appraisal) calculate or estimate every individualadaptability x(t)→.Step 3 (colony evolution).Step 4 (termination examination) if →x (t +1)satisfies the terminationcriterion, then outputs the individual →x (t +1) which has the biggestadaptability to take the optimal solution , termination computation;Otherwiset ← t+1, then go to step 2.SEC has its essential merit, including take optimizes the variable the geneticcode as the operation, the search object.It only applies "the adaption value" theinformation, but do not have to apply concrete value of the objective functionand other auxiliary informations. The non-simple point operation, the usecommunity searches the strategy. It use probability searches the mechanism.Summariy, SEC have the following merits such as general, parallel, steady,simple and the global optimization ability.Second chapter main abstractly indicates the formalized model of thesimulation evolution algorithm (SEA), including the choice algorithm, theoverlapping operator, the variation operator and the genetic code and so on theessential factor, as well as in the algorithm execution involves all kinds ofparameter.Evolution algorithm SEA provide general frame,It taking advantage ofexplained the biological evolution and a genetic variation mechanism solutionquestion, but if had effectively to use, specially aimed at each kind of differentapplication goal but to achieve "with a clear goal" then needed to have thecertain skill.Third chapter mainly narrated the simulation evolution computation to carry outthe skill.These methods including the outstanding recording and the "fathersand sons to mix " the choice strategy, the adaption value sharing strategy, theparallel realization strategy. The adaption value sharing strategy main thought isbetween the colony similar individual the compatibility took the sharedresource, it use individual the similar individual number and the similar degreereadjusts its adaptbility in the colony, by achieves both forms the microhabitatevolution environment in the similar individual, and suppresses similarindividual gradually in the whole evolution the reproduction number, bymaintains the colony multiple-goal. The parallel realization strategy, includingthe parallel strategy based on the colony grouping and the parallel realizationstrategy based on the search space.Fourth chapter is mainly the search mechanism of genetic algorithms, its basictheory question mainly involves the following three aspects: First, relatedgenetic algorithms search mechanism, second, related algorithm characteranalysis, third, related algorithm complex analysis.The colony growth equation describes the genetic algorithms searchprocess, which individuals are easy to survive or easy to washed out, thisprocess is one kind of theory and the algorithm, it by Holland et al. proposesand development. The people thought that, The genetic algorithms solutionprocess is not one by one test each gene enumerates the combination in thesearch space, but is through some good patterns, like the picture builds thebuilding block to be same, splices them, thus the gradual structure leaves theadaptability more and more high individual code string. This cognition is calledit the gene block supposition. The pattern theorem and the gene blocksupposition is the early time the explanation for searches the mechanism to thegenetic algorithms.We analyzes the genetic algorithms search mechanism content and drawthe following conclusion: 1, the genetic algorithms are defined by intersect andchoice in the colony iterative process which the variation operation repetition. 2,if does not have the variation, then the genetic algorithms accurate search mayreach the territory is the minimum pattern which the initial colony decided.Therefore, this time genetic algorithms can find the question the overallsituation optimal solution when also if and only if when contains the optimalsolution. 3, if the choice and the overlapping operator are the partial searchoperator, then the variation operator may think the overall search operator(because his function lies in expanded search space, enables genetic algorithmscan search in the entire spatial).Fifth chapter main introduce genetic algorithms each astringent definition,representative astringency analysis method (model) and correlation geneticalgorithms astringent result.The genetic algorithms are repetition choice unceasingly, overlapping andthe variation process, each kind of genetic operation only is all related with thecurrent condition, but had nothing to do with with the before condition,therefore the genetic algorithms may describe -the Markov chain, thus itsastringency might perform using the Markov chain theory to study.The axiomization method is the axiomization description which is basedon to simulates the evolution operation, and use each correlation operationessential characteristic number to the algorithm astringency to make theprobability estimate directly. This method not only is suitable includesnon-homogeneous self-adapted and so on each kind of genetic algorithmsanalysis, moreover analysis method itself direct-viewing, is clear, deduct theconclusion also provides the reference to the genetic algorithms kinds ofparameter choice.Sixth chapter is the genetic algorithms practical application in the handback vein image segement. The human body vein blood vessel near-infraredimage formation is an item applies extremely widespread technology in themedicine, this technology takes one new biometrics identification technology.This step is prepared to extract the venous blood pipe-line lately. Using thetraditional image valve value segement method when carries on analysis anddeal with to the image, mostly is according to the image grey level histogramcarries on the valve value segement, this method has a big shortcoming whichdose not overcome the image’s big noises. This article uses based on the geneticalgorithms territory value segement technology, this method has the goodlocalization characteristic in the time domain and the frequency range, it canautomatic selection threshold under the different criterion, thus can enormousremove the noise, obtains the good segement effect.Based on genetic algorithms two-dimensional entropy territory valuesegement method as follows: (1) stochastically produces N individuales, formsthe initial chromosome colony;(2) calculates in the colony each chromosomeadaptability;(3) calculates the average adaptability of the colony, eachchromosome adaptability normalization;(4)according to normalized theadaptability value, bestowal on survival probability to each chromosome,according to this probability choice n (n<<N)the chromosome;(5) to n whichselects to the chromosome carries on overlapping and the variation operation,forms 2n innovate chromosomes, the overlapping operation and the variationoperation position are both stochastically produces;(6) with a 2n chromosomeswhich produces newly replaces the 2n chromosomes of adaptability, forms thenew colony;(7) if satisfies the termination condition, then the code adaptabilitybiggest chromosome, obtains the biggest chromosome, obtains the question thesolution, otherwise returns to (2). The experiment indicated our method isfeasible.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile