Research On Cloud Model Theory And Its Application In Clustering Analysis Of Color Image

Cloud model is a class of mapping with uncertainty that is used to model the relationshipbetween the qualitative concepts represented by natural language and the quantitative datumrepresenting such a qualitative concept based on the probability theory and fuzzy mathematics.It models the fuzziness by employing the objective data that denote the features of theconsidered thing. Meanwhile, cloud model can implement the bidirectional transformationbetween qualitative concepts and corresponding quantitative datum, and considers thefuzziness and randomness together during the process to solve the problems with uncertainty.This paper is devoted to some fundamental problem of cloud model theory. The proposedresults first analyzed the atomized property of cloud model and the mathematical property ofcloud distribution in detail. On this basis, the remainder investigated the backward cloudmodel algorithm, cloud mixture model and their applications in color image clusteringanalysis. The main work of this paper is as follows.(1) The ratio of entropy to hyper entropy is defined as atomized factor to measure thedisperse degree of cloud drops. This proposed method overcome the defect of the existingresults that did not consider the constraint of the cloud distribution probability densityfunction on the relation of entropy and hyper entropy and thus can not effective measure thedisperse degree of cloud drops. On this basis, the influence of atomized factor to the clouddrops distribution is discussed from following aspects, numerical characteristic of clouddistribution; disperse degree of cloud drops and the cloud drop density.(2) A novel algorithm of backward cloud model without certainty is presented based onthe property of the fourth moment of cloud distribution. Some comparative experiments arecarried out under the condition that the value of atomized factor and the numbers of clouddrops are different. The results indicate that the proposed algorithm is superior to the existingones in accuracy and stability.(3) Inspired by the theory of the gaussian mixture model, cloud mixture model (CMM) isproposed to solve the fuzzy clustering problems among data clusters based on the fact that thecloud drop quantitative data follows the cloud distribution, and the property that thequantitative datum can be transformed as the qualitative concepts and vice versa. Thepresented algorithm is a new model method in this field. Thereafter, this algorithm is used todeal with the clustering analysis of color image. The results of experiment demonstrate thatCMM has the better clustering performance than GMM, FCM and K-means, and extend theapplication of cloud model theory in pattern recognition.(4) A semi-supervised clustering analysis algorithm is proposed by employing theinformation granulating theory of cloud model and the two-dimension cloud model. By using this algorithm in the vision system of multiple biomimetic robot fish platform, the results ofexperiment show that it is robust to the uncertainties of environmental illumination and cansegment the objective accurately.