Seminar Series

The influence of image complexity on perceptual processing

Natural images widely differ in terms of their complexity. Some images only consist of a clear sky with a bird in the foreground while other images are crowded with leaves, trees and/or people. One method to index this complexity is by looking at the contrast energy (CE) and the spatial coherency (SC) of an image, which together describe the compression ratio of an image (Scholte et al., 2009). The SC of an image describe an axis that on one contains images with a power-law distribution that covaries with attributes such as the presence of a clear figure/ground segmentation and the presence of manmade objects and on the other with end a Gaussian distribution that covaries with fractionation and natural scene elements. Here we present a series of new experiments in which we ask subjects to indicate whether an animal is present or absent. If subject receive a cue they should inhibit their response (Stop Task). Images either have a low, medium or high SC value. Drift diffusion analysis shows that subjects accumulate information slower in high SC condition in combination with a lower decision boundary. Responses for target absent trials are slower and coincide with a higher decision boundary. This indicates that the SC of an image indexes the degree to which detecting an animal is in effectively a pop-out task (lower SC values) vs. a slower visual search task (high SC values). Using BOLD-MRI we also find substantially different connectivity profiles between areas involved in behavioral control and areas involved in visual processing depending on the SC values of the images. Finally we find activity in area V1 for the difference between the animal and non-animal scenes if the SC value is high and not if the SC value is low. ERP's of the animal and non-animal stimuli indicate this differences occurs after 200 ms. In sum we see that detecting animals in low SC scenes is similar to a pop-out task and appears to involve mainly feed-forward processing, this task in high SC scenes is more akin to a visual search task an involved recurrent processing. Together the results indicate that SC is a good proxy for image complexity and the processing style of the visual information, irrespective of task.