K-means clustering [1] has been widely used in various applications. One intrinsic limitation in K-means clustering is that the choice of initial clustering centroids may highly influence the performance of the algorithm. Some existing K-means initialization algorithms could generally achieve good results. However, in certain cases, such as CT images that contain several materials with similar gray-levels, such existing initialization algorithms will lead to poor performance in distinguishing those materials. We propose an automatic K-means initialization algorithm based on histogram analysis, which manages to overcome the aforementioned deficiency. Results demonstrate that our method achieves high efficiency in terms of finding starting points close to ground truth so that offers reliable segmentation results for CT images in aforementioned situation.
CITATION STYLE
Tian, M., Yang, Q., Maier, A., Schasiepen, I., Maass, N., & Elter, M. (2013). Automatic histogram-based initialization of K-means clustering in CT. In Informatik aktuell (pp. 277–282). Kluwer Academic Publishers. https://doi.org/10.1007/978-3-642-36480-8_49
Mendeley helps you to discover research relevant for your work.