Image saliency detection based on rectangular-wave spectrum analysis

The saliency detection is an important step in many computer graphic and multi-media technology tasks because of its wide applicability. Computer graphic tasks (such as image segmentation, image labeling and tracking) rely mainly on fine saliency maps. However, most of the existing methods are defective due to their poor ability in generating fine boundary or high recall rate. This paper proposes a saliency detection method. The given image is divided into several regions which convolute with a pre-set rectangle-wave template later. For each region, the final saliency value is determined by the difference between themselves and the regions next to them as well as their uniqueness when they are compared with the whole images according to the corresponding response. Repeated tests on the different data sets prove its high precision-recall rate. Additionally, the boundary in our saliency map is clearer and finer.