Image Segmentation with Clustering K-Means and Watershed Transform

Image segmentation is a very important process for multimedia applications. Multimedia databases use segmentation for the storage and indexing of images. This paper presents a way to segment images by applying both a clustering method and watershed transformation. It is well known that the major drawback of the watershed transformation method is the oversegmentation phenomenon it produces. For this reason the image is first segmented with the K-Means clustering method. Another well-known fact is that after applying the K-Means algorithm the output image contains a lot of noise. This is why the image is then filtered with a Gasussian blur filter. Finally the watershed transformation is applied. Tests results obtained using the images from a segmentation evaluation database, show that using this particular combination of methods results in a highly reduced oversegmentation.