Predicting performance in natural scene searches

Completely natural scene search is a paradigm that cannot be directly compared to the typical types of search task studied, where objects are distinct and definable. Here we have look at the possibility of predicting the performance of humans for completely natural scene tasks, using a direct comparison of human performance against new and existing computer models of viewing natural images. For the human task, participants were asked to perform a target present/target absent search task on 120 natural Scenes, the target being a subsection of the Scene and the false-target matched to the scene. The identical task was given to a selection of reproductions of existing computer processing techniques, including Feature congestion (Rosenholtz et al., 2005 SIGCHI 761–770), Saliency (Itti & Koch, 2001 Journal of Electronic Imaging 10 161–169), Target Acquisition Model (Zelinsky, 2008 Psychological Review 115 787–835) and a new variation on the Visual Difference Predictor (To et al., 2008 Proceedings of the Royal Society B: Biological Sciences 275 2299–2308). We show that the models are very bad at generating parameters that predict performance, but that A' of Human performance is predicted pretty well by the simple clutter in the image and these results lead us to conclude that in natural search tasks, the nature of both the Scene and the Target are important, and that the global influence of local feature groups can have an influence of the task difficulty.