Models & Optimisation and Mathematical Analysis Journal
Volume 6, Numéro 1, Pages 28-33
2018-12-27

K-means Clustering Based On Multiresolution Images And Spatial Constraints

Authors : Benzian Mohamed Yaghmorasan . Benamrane Nacéra .

Abstract

This paper proposes a new K-means segmentation approach applied on multiresolution images and based on spatial constraints. K-means clustering is performed first on various image resolution levels where the limit of resolution level can reach 1/8th of image. Then, clustering result of each pixel p of the original image can be updated depending on clustering result k on lower resolution images and on presence ratio of k in the spatial neighbourhood of p in the original image. Image analysis at lower resolution allows having rough clustering result which means that a pixel is affected to a cluster to which the majority of its pixel neighbourhood belongs. The aim of this approach is to minimize clustering errors depending on the spatial cluster repartition at the neighbourhood of each pixel in order to get more homogeneous regions and eliminate noisy regions in the image. The approach is tested on simple and medical images by adding a gaussian noise and varying resolution level for a better analysis. The results of multiresolution clustering are satisfactory and a comparison is made with standard K-means.

Keywords

Segmentation ; K-means ; Multiresolution ; Gaussian Noise ; Spatial Constraints ; Clustering