A new FPCA-based fast segmentation method for color images

Fuzzy objective function-based clustering methods are proved to be fast tools for classification and segmentation purposes. Unfortunately, most of the available fuzzy clustering methods are using the spherical or ellipsoidal distances, which are proved to result in spurious clusters, when working on color data. In this paper, a general case of clustering is discussed and a general method is proposed and its convergence is proved. Also, it is proved that the FCM and the FCV methods are special cases of the proposed method. Based on the general method, a special case for color image processing is proposed. The clustering method is based on a likelihood measure, and is proved to outperform the Euclidean and the Mahalanobis distances, in color fields. Based on the proposed color clustering method, a new fast fuzzy segmentation method is proposed and is proved to be highly efficient Comparison of the results with the FCM, proves the superiority of the proposed segmentation method.