A comparative evaluation of similarity measurement algorithms within a colour palette

Recently, there has been interest in the development of colour palettes from images. Colour palettes have long been used by designers to communicate colours and their relationships but increasingly palettes are being derived automatically from digital images, concepts, or from a plethora of digital design tools online. Methods to predict differences between palettes are growing in popularity. This study is concerned with the prediction of visual self-similarity for colour palettes with large numbers of patches. A psychophysical experiment was carried out to collect the human judgments of similarity and then six different algorithms were introduced and evaluated in terms of their ability to predict the psychophysical data. Two methods to quantify the agreement between the visual data and algorithm predictions were used based on regression analysis with coefficient of determination for the goodness of fit and multidimensional scaling with loss function Kruskal's stress. Of the six algorithms, the Pearson correlation coefficient method was considered to give the best performance.