Efficient use of extrinsic cues for salient object detection

Determine efficient strategies to exploit extrinsic cues—including ground-truth training annotations, similar-image retrieval, and saliency co-occurrence across image sets—for salient object detection across different application scenarios.

Background

The survey categorizes methods by intrinsic versus extrinsic cues. Extrinsic cues encompass using annotated training data (supervised learning), leveraging similar images retrieved from large corpora, and exploiting co-occurrence of saliency across sets of images (co-saliency).

While intrinsic cues have been widely explored, the authors note that strategies for incorporating these extrinsic cues effectively and efficiently in diverse applications remain unsettled, highlighting a need for principled methods to integrate such information.