Research On Color Image Segmentation Algorithm Based On Suppressed Possibilistic Fuzzy Clustering

Partition clustering is an important unsupervised pattern recognition method.Possibilistic fuzzy c-means clustering(PFCM)algorithm combines fuzzy membership and possibilistic membership,which can partly inherit the stability of fuzzy c-means clustering(FCM)algorithm and the noise robustness of possibilistic c-means clustering(PCM)algorithm,so it has attracted much attention.However,there are several problems in the possibilistic fuzzy c-means clustering as follows.Firstly,the Euclidean distance used in the PFCM treats all features of sample equally,which leads to the imbalance between sample features being ignored.Secondly,due to the difficulty of the PFCM in membership-weight parameter setting and the lack of the between-class relationships in possibilistic memberships,the PFCM always produces significant center deviations and overlapping centers for multi-class datasets with strong noise injection.In order to solve the above problems,this thesis introduces the “suppressed competitive learning” mechanism,the feature-weighted method and shadow set theory into the possibilistic fuzzy c-means clustering algorithm,and two kinds of suppressed possibilistic fuzzy clustering algorithms are proposed: a feature-weighted suppressed possibilistic fuzzy c-means clustering(FW-SPFCM)algorithm and a suppressed possibilistic fuzzy c-means clustering based on shadow set(SS-PFCM)algorithm.Then this thesis applies the two algorithms to color image segmentation to improve the segmentation results of noisy images.The main work of this thesis is as follows:(1)To improve the clustering effect of the imbalance between sample features on multiclass clustering with noise injection,a suppressed possibilistic fuzzy c-means clustering(FW-S-PFCM)algorithm is proposed.Firstly,this algorithm introduces a feature-weighted parameter to automatically assign feature-weight values to different features and different clusters according to the distribution of samples,thus overcoming the influence of feature imbalance and improving clustering effects for multi-dimensional datasets.Secondly,combined with the feature-weight matrix,a “suppressed competitive learning” strategy is designed to resolve the center-overlapping problem in noisy multi-class dataset clustering caused by the lack of between-class relationships in the PFCM algorithm and improved the anti-noise ability of the PFCM algorithm.(2)To improve the segmentation results of color images with unbalanced color features and strong noise pollution,a noise image segmentation method is designed based on the proposed FW-S-PFCM algorithm,which combines the strong robustness to long-distance noise and the noise-identification ability of possibilistic memberships.Firstly,the FW-SPFCM is used to automatically obtain the feature-weight values,so as to overcome the influence of unbalanced features of color images.Secondly,an automatic identification method of salt-and-pepper noise(singular value)and a neighborhood label replacement method are designed to post-process the segmented image by using the absolute attribute of the possibilistic membership in the PFCM,so as to improve the segmentation effect of the color image polluted by salt-and-pepper noise.(3)In order to further overcome the problems of center deviations and overlapping centers for multi-class datasets with strong noise injection,a suppressed possibilistic fuzzy c-means clustering based on shadow set(SS-PFCM)is proposed by introducing the shadow set theory and the “suppressed competitive learning” mechanism.Firstly,the possibilistic membership distribution characteristics of samples are statistically analyzed by using the shadow set theory,and each type of data is adaptively divided into three approximate regions(inner-core data region,edge region and outer-core noise region).Then,the suppressed learning strategy is designed according to the possibilistic membership corresponding to the data of three approximate regions to overcome the coincident clustering problem.The“suppressed competitive learning” mechanism is designed according to the fuzzy membership degree to reduce the iteration times of the PFCM.Finally,the SS-PFCM algorithm is applied to color image segmentation,and the noise is automatically identified and removed by combining the shadow set region division scheme,so as to improve the segmentation effect of color images with salt-and-pepper noise pollution.