| Image segmentation is a very critical step in the process of imageprocessing, segmentation result directly influences the quality of imagerecognition and analysis, so the image segmentation algorithm research hasbeen paid more and more attention. The traditional segmentation methodincluding threshold segmentation, region segmentation and edge detectionmethod, etc. In recent years, some segmentation method based on the newtheory has began to enter the line of sight of people, such as: segmentationmethod based on mathematical morphology, segmentation method based onneural network, the segmentation method based on fuzzy control andsegmentation method based on genetic algorithm, etc.Cloud model is proposed by the theory of probability theory and fuzzyset theory. It is a transformation between the qualitative concept andquantitative expression. Normal cloud model is a kind of basic cloud model, alot of qualitative concept in the natural world can be approximately regarded asnormal or half of the normal distribution, so many problems can be seen as anormal cloud model to solve. Cloud droplets is uncertain, the same clouddroplets can belong to different cloud, and the different of cloud droplets can also belonging to the same cloud. This characteristic can be "soft" partition ofuncertain information. Image segmentation is a process that classified the pixelsto extract the target area on the basis of the known information. So theuncertainty of cloud droplets can be applied in the field of image segmentation.The application of cloud model in image segmentation has been a lot ofresearch and achieved good results. But in the field of new theory and newtechnology, the improved research is relatively less. This paper to improve thestudy from two fields: mathematical morphology and genetic algorithm.In the field of image segmentation based on mathematical morphology, themost typical segmentation method is watershed algorithm. The main problemwatershed algorithm in the image segmentation is prone to over-segmentationphenomenon, so through the cloud model integrated algorithm to merge areascan successfully solve this problem. When making combine with cloud modelneed to merge small clouds. What kind of cloud is fit to merger? The traditionalalgorithm did not give a simple and effective method. In order to solve theproblem, this paper put forward the concept of cloud close degree, which isdefined from the point of view of relative distance and cross section size. Thismethod made the results more reasonable. Using cloud synthesis algorithm withthe determination methods in watershed algorithm can get better segmentationeffect.In the field of image segmentation based on genetic algorithm, the valueset of crossover probability and mutation probability directly affect the search results of the algorithm. The traditional genetic algorithm is easy to search intolocal optimum because the value of crossover probability and mutationprobability is certain. Adaptive genetic algorithm can adaptively adjust thecrossover and mutation probability, but the problem of genetic algorithm easilyto fall into local optimal problem has not been effectively resolved. This articleuse cloud model to solve this problem. Using X condition cloud generator toreconstruct the genetic crossover probability and mutation probability, whatmakes the algorithm can be adaptive adjustment, and effectively avoid theproblem of easily fall into local optimal. By using this method, the algorithmcan find the best segmentation threshold quickly and steady. |
