ANÁLISE COMPARATIVA ENTRE K-MEANS, FCM E ALGORITMOS CKMEANS PARA SEGMENTAÇÃO DE IMAGEM

COMPARATIVE ANALYSIS BETWEEN K-MEANS, FCM, AND CKMEANS ALGORITHMS FOR IMAGE SEGMENTATION

Resumen

1. Clustering algorithms are often used for image segmentation, aiming to group pixels by their similarity and uniformity. This process is useful to detect and highlight important areas of an image, making its analysis easier in several applications such as remote sensing and medical diagnosis. This paper have the main objective to compare the K-Means hard clustering algorithm to the FCM and ckMeans fuzzy clustering algorithms in image segmentation applications, using the R statistical programming language for analysis and visualization of the results. Uncertainty in the clustering process is discussed via the use of the alpha-cut parameter. Two experiments were conducted, using an image from an open database and an aerial image of a Catarinense city. It was found that the three methods produced similar results, when crisp clusters were considered. Fuzzy membership results of FCM and ckMeans were also compared, and it was found that, although very similar, ckMeans produced slightly lower levels of uncertainty than FCM. It was found that K-Means presents the best computational performance among the algorithms compared, which is expected due to its crisp nature. Among the fuzzy algorithms compared, ckMeans presented better performance, and FCM required less memory.