Objective metric for colour image comparison

In this paper, a method for colour image comparison is proposed. An indexation step is performed using a Kohonen neural network, with the SOM (Self Organizing Map) algorithm. Therefore, a colour image in the Red, Green, Blue (RGB) space is interpreted as a 256 colour image. The comparison step, between two indexed images, is based on a global dissimilarity measure, which is an extension of Baddeley's distance, adapted to this colour point set. Experimental results obtained with real images are presented.