Segmentation of Natural Images using Self-Organising Feature Maps

We propose a technique for the segmentation of images using both colour and texture information. This information is in the form of lowpass colour responses and sixteen Gabor texture lter responses. A self-organising feature map is trained to label any set of responses into appropriate categories. We quantify the success of the approach and assess the importance of colour and texture in the segmentation process. The technique is shown to be highly successful in segmenting both arti cial and natural outdoor scenes.