A novel measure of estimating colour purity of image regions

The estimation of colour purity of image regions is very important for several image processing analyses such as measuring image viewability, constraining the region growing type image segmentation algorithms, quantifying the quality of image segmentation, and estimating object colour homogeneity. The most commonly used image histogram measure of colour purity is to compute variability of the red, green and blue channels respectively. Unfortunately, this measure is not a reliable indicator of colour purity since two very different histograms can yield the same measurements. We propose a novel histogram based measurement of colour purity that is computed by clustering histogram data. The measurement is dependent on the size of clusters found as well as the distance between them. We show that for explosive detection application using dual energy colour coded images, the proposed measure correlates much better with human judgment of colour purity compared to the colour variance of the three colour channels.