Color image segmentation using a semi-wrapped gaussian mixture model

Abstract

This article deals with color image segmentation in the hue-saturation-value space. Hue, saturation and value components are samples on a cylinder. A model for such data is provided by the semi-wrapped Gaussian distribution. Further its mixture is used to approximate the hue-saturation-value distribution. The mixture parameters are estimated using the standard EM algorithm. The results are obtained on Berkeley segmentation dataset. Comparisons are made with vM-Gauss mixture model, GMM and Mean-Shift procedures. Experimental results reveal improvement in segmentation by our method. © 2011 Springer-Verlag Berlin Heidelberg.