Colour map design through optimization

Abstract

Often in engineering systems, full-colour images have to be displayed on limited hardware, for example on mobile devices or embedded systems that can only handle a limited number of colours. Therefore an image is converted into an indexed map from where the indices point to specific colours in a fixed-size colour map generated for that image. The choice of an optimal colour map, or palette, is therefore crucial as it directly determines the quality of the resulting image. Typically, standard quantization algorithms are used to create colour maps. Whereas these algorithms employ domain specific knowledge, in this work a variant of simulated annealing (SA) was employed as a standard black-box optimization algorithm for colour map generation. The main advantage of black-box optimization algorithms is that they do not require any domain specific knowledge yet are able to provide a near optimal solution. The effectiveness of the approach is evaluated by comparing its performance with several specialized colour quantization algorithms. The results obtained show that even without any domain specific knowledge the SA based algorithm is able to outperform standard quantization algorithms. To further improve the performance of the algorithm the SA technique was combined with a standard k-means clustering technique. This hybrid quantization algorithm is shown to outperform all other algorithms and hence to provide images with superior image quality.