Uniform color spaces clustering in an unsupervised manner

Color quantization is a critical task, frequently involved in image processing that reduces the number of distinct colors used in an image while retaining as much of the original representation capabilities. The key aspect here is to find the optimal palette and evaluate against unprocessed target images. The purpose of this paper is to compare the effectiveness of three well known unsupervised vector quantization algorithms (Neural Gas, Growing Neural Gas and Instantaneous Topological Map) in the field of color abstraction. Evaluation data for L*a*b* and L*u*v* uniform color spaces and a number of quality indices, exhibiting the performance in terms of overall quality, are presented.