A comparison of biologically-inspired methods for unsupervised salient object detection

The detection of salient objects in images is a challenging problem with many potential applications. In this work, the applicability of computational models of visual attention to the detection of salient objects within images in an unsupervised manner is considered. The classic Itti-Koch-Niebur model, Graph-Based Visual Saliency (GBVS) model, and Image Signature are compared. Results show that all three methods can effectively be used to detect objects. The Itti-Koch-Niebur and GBVS models perform best, but the Image Signature is also shown to provide effective, unsupervised salient object detection. The methods developed in this work can be used to evaluate other unsupervised methods of object detection.