Top-down control of visual attention in object detection

Abstract

Current computational models of visual attention focus on bottom-up information and ignore scene context. However, studies in visual cognition show that humans use context to facilitate object detection in natural scenes by directing their attention or eyes to diagnostic regions. Here we propose a model of attention guidance based on global scene configuration. We show that the statistics of low-level features across the scene image determine where a specific object (e.g. a person) should be located. Human eye movements show that regions chosen by the top-down model agree with regions scrutinized by human observers performing a visual search task for people. The results validate the proposition that top-down information from visual context modulates the saliency of image regions during the task of object detection. Contextual information provides a shortcut for efficient object detection systems.

The importance of information localization in scene gist recognition.

“…The term gist is not always clearly defined (though see Oliva, 2005) but is most frequently operationalized as the scene's basic level category, for example, "beach" or "street" (Tversky & Hemenway, 1983), and we follow that convention here. Gist information appears to guide viewers' inspection of the scene (Loftus & Mackworth, 1978;Oliva, Torralba, Castelhano, & Henderson, 2003), may aid object recognition in the scene (Boyce & Pollatsek, 1992;Davenport & Potter, 2004;De Graef, De Troy, & D'Ydewalle, 1992;Hollingworth & Henderson, 1998;Palmer, 1975), and affects later memory of the scene (Brewer & Treyens, 1981;Pezdek, Whetstone, Reynolds, Askari, & Dougherty, 1989). Given the speed of gist perception, the information underlying gist recognition may be based on holistic, low-level scene properties (Oliva & Torralba, 2001;Renninger & Malik, 2004;Vailaya, Jain, & Zhang, 1998), rather than based on detecting or recognizing individual objects (cf.…”

An Attentional System Combining Top-Down and Bottom-Up Influences

“…This makes the combination rigid and non-flexible, which may result in loss of important bottom-up information. Oliva et al [1] show that top-down information from visual context modulates the saliency of image regions during the task of object detection. Their model learns the relationship between context features and the location of the target during past experience in order to select interesting regions of the image.…”

Learning a Saliency Map for Fixation Prediction

“…A large amount of previous work concentrates on bottom-up saliency detection [1]- [4]. Top-down attention guides gaze in a task-dependent and goal-driven manner, which is dominated by object information [5]. In the last few decades, extensive research has been conducted to predict visual attention.…”