Automatic colour segmentation and colour palette identification of complex images

Abstract

A fast and simple colour classification algorithm was developed based on k-means++ algorithm for quick colour analysis of complex images. The algorithm processes Gaussian blurred CIELAB version of the original image to segment the scene in clusters of six colours with the highest representation in the image. It identifies the colours with the help of the ISCC-NBS colour terminology and locates them on the scene. The accuracy of our algorithm was evaluated with a psycho-visual experiment. The results show a dominant colour classification coherent with human visual perception. The algorithm is efficient for quick analysis of complex colour images to retrieve automatically the colour composition. It can be applied directly in various domains for example: lamp testing for colour distortion, image retrieval, artwork analysis etc.