Using Machines to Improve Human Saliency Detection

Abstract

Humans are adept at identifying informative regions in individual images, but it is a slow and often tedious task to identify the salient parts of every image in a large corpus. A machine, on the other hand, can sift through a large amount of data quickly, but machine methods for identifying salient regions are unreliable. In this work, we develop a new method for identifying salient regions in images and compare this to two previously reported approaches. We then consider how such machine-saliency methods can be used to improve human performance in a realistic target-detection task.