Study Of Lmage Segmentation Methods Based On Fuzzy Lnformation Processing

ANSTaxCTImage segInentation is an importan and basic technique in computer vision. It is a keytechnique in Anage understanding, boaging eqet recognition and tracking, robot vision, etc.It will make a strong bopact on vision syStem performance. Being an illposed problem ofinformation shortcoming, it is necessare to aPply reStraint conditions and premise AnowIedgeto regUlM driage segInentaion problem. Considering of diversity and comPlexity of visiontaSks and variety of information obtained from hage, constraint conditions are different fordifferent taSk. Due to ignorance of information loss in boaging process, it is difficult todescribe Prffoise lcnowledge accurately and mathematically Moreovef, because of unknownreasons caused by boaging process, there are lilielthood and uncertainty betWeen mpt andbackground. Fop driation processing tecboque developed frOm tw set theory haveadvantage to process uncertainty event and describe uncertainty Anowledge. Imagesegmentation is an intelligent process to partition driage pirel based on boage informationand premise constraint conditions.In this dissertaion, linage segInentation technique using different image information basedon fhazy infOrmation processing method is stUdied, and a series of new ideas and approachesare presented, with good effects. This dissertation is classified seven chaPters, which isorganized as fOllows:In chaPter l, a brief review of the history and the research statos in computer vision ispresented. It is illustrated that boage segmentation is an hotspot problem which hasperplexed development of comPuter vision field within several decade, and then problemswhich image segInentation itSelf exist are discussed, and point out, that it is necessary to usefhazy information processing techniques at hoage segmentation. Finally, the mainachievements of this dissertaion are sununarized.In chaPter 2, basic principle of Anage segInentation is illustrated. And relationshipbetWeen fhazy cIassification techniques and driage segInentaion is discussed. Since Anagesegmentation is an intelligent partition process, this chapter emphasis on researching fhazyclassification techniques which to be aPpIied in driage segInentation. Thre newclassification aPproaches are given: the first is a new tw soft cIustering method basedhuman partition chacteristicsectional set FCM clusteT, the second is a new patialweighted FCM cluster method by POtential driction, and the third is a new non-iterationpotential driction clustering method. Being a non-linear oPthoal problem, Cluster is verysensitive to initialhation, so initializing problem of cluster is stUdied, and a new fimctionalinitialhation method is given. In process of classification, it is a key for intelligence PrOCessto adaptively determine class number of sample data set. In this chapter, a new cluster validly method based on fuzzy neighborhood measure is presented, which is used to adaptive determine class number of data setIn chapter 3, multi-thresholds image segmentation method based on image gray statistic information are studied, and two effective multi-threshold approaches: potential function clustering adaptive multi-thresholding, histogram fuzzy constraint cluster adaptive multitbresholding are presented. These approaches are analyzed deeply, and corresponding experiments are given in detail, with satisfaction effect.In chapter 4, by enhancing peak-valley feature of image histogram, two segmentation methods are proposed, one of which is fuzzy adaptive histogram enhancing segmentation based on region smoothness measure. Other is color image segmentation based on hue histogram fuzzy enhancing, with human抯 common sense to color.In chapter 5, image fuzzy segmentation approaches based spatial and gray informations are studied. Image infonnation entropy is discussed. According to different characteristic of image, three segmentation methods: fuzzy region grow segmentation with heuristic knowledge~ pyramid fuzzy cl...