The Statistical Saliency Model Can Choose Colors for Items on Maps

作     者：Shive, Joshua Rosichan, Sharra Davis, Sherika Wade, Christena Ellison, James Santoni-Sanchez, Santos 

作者机构：Tennessee State Univ Dept Psychol Nashville TN 37209 USA Univ Tennessee Sch Informat Sci Knoxville TN 37996 USA Middle Tennessee State Univ Dept Psychol Murfreesboro TN 37132 USA 

出 版 物：《IEEE TRANSACTIONS ON HUMAN-MACHINE SYSTEMS》 (IEEE Trans. Human Mach. Syst.)

年 卷 期：2019年第49卷第6期

页      面：569-578页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Science Foundation Office of Naval Research 

主　　题：Algorithm design and analysis Visualization Clutter Color Predictive models Computational modeling Display clutter display design human-computer interface human performance modeling information visualization visual search scanning 

摘      要：We show how a model of visual salience that was originally developed to explain human visual search performance can suggest display design choices that reduce search time for items. The statistical saliency model proposes that the time to find an item on a visual display depends on the similarity between a target items features and the statistical distribution of display features. In the present study, observers rated the amount of display clutter on a set of MapQuest maps containing colored pushpins. We identified a group of high-clutter maps and a group of low-clutter maps. Next, we used the statistical saliency model to choose colors for new pushpins placed on those maps. We show that the models color assignments depend on the colors the display contains. Map designs produced using this method were tested in a visual search experiment. Search time decreased as a pushpins predicted salience increased. In addition, choosing low salience colors led to slower search times for items on high-clutter displays than for items on low-clutter displays. The method we describe works with real images and does not require any parameter fitting. This study provides evidence that computational models of visual perception have potential as display design tools.