Multi-resolution learning vector quantisation based automatic colour clustering

Image segmentation is a challenging problem in computer vision. Feature clustering based image segmentation schemes are extensively researched topics in recent years. Particularly, colour clustering schemes are being widely applied for motion detection and tracking applications. In this paper we present a novel colour clustering methodology based on vector quantisation. The proposed method applies a learning vector quantisation approach with multi-scale image hierarchy to colour clustering using the hue, saturation, value (HSV) colour space model in order to obtain robust colour image segmentation. Results from experiments are presented, including a comparative analysis with the c-means algorithm.