Color Image Clutter Metrics for Predicting Human Target Acquisition Performance

A method to quantify clutter in color images is presented. This method is based on a grayscale clutter metric, target structure similarity (TSSIM) metric, which showed the best correlation with human perception than the other available metrics. The proposed new metric is designed to measure the background clutter in the perceptually decorrelated quasi-uniform color space of CIELAB. The TSSIM metric is extended to include color by applying it to the individual dimensions of the CIELAB and combining the individual components together using weighted means. To evaluate the new metric, TSSIM and this new metric are both applied to the Search_2 dataset, a visible imagery set with perception experiment data. Experimental results show the new metric correlates perceptual data better than the TSSIM metric and point the way to various improvements.