ABSTRACT:

This research argues that visual information is weighted by prior expectations before saccadic decisions are made. Expectations regarding the location of objects in natural scenes heavily influence saccadic decisions and are based on experiences with features of those scenes. A database of natural scenes is collected and features extracted. This information is then compared with behavioral data of humans performing a search task to determine the relative influences of the features.

The present study investigates the role of scene context in visual search and observers’ prior expectations of target locations employing stimuli comprised of natural scenes. The work involves several components. Two of these components, determining expected locations of objects and predicting saccadic eye movements, are primarily the responsibility of Barbara Drescher.

Laura Boucheron compares current state of the art in semantic image search and contextual priming algorithms with the behavioral data with hopes of providing a means to improve their performance. In addition to developing methods of extracting spatial configuration of high-level objects, it is possible that this research could also improve relevance feedback in data-mining and image search applications as well as provide insight into behavioral patterns.