KSLIC: K-mediods Clustering Based Simple Linear Iterative Clustering

Abstract

Simple Linear Iterative Clustering (SLIC) is one of the most excellent superpixel segmentation algorithms with the most comprehensive performance and is widely used in various scenes of production and living. As a preprocessing step in image processing, superpixel segmentation should meet various demands in real life as much as possible, but SLIC is highly sensitive to noise. In this paper, a K-mediods clustering based simple linear iterative clustering (KSLIC) is proposed, which replaces the K-means clustering in SLIC with a modified local K-mediods clustering. To evaluate the performance of KSLIC, we test it on BSD500 benchmark dataset. The results show that it outperforms SLIC in terms of different noise environments including Gaussian noise, multiplicative noise and salt and pepper noise.