| Image saliency detection is an important research branch in computer vision area with multiple applications. It can be the preprocessing step of image, obtaining the regions of interest and removing redundant information in the image, which can improve the efficiency of image processing. However, there exist various challenges in detecting process, such as the complex background of the image, the varies target and so on, which makes it still a challenging topic to develop a robust detection method.In this paper, we propose a bottom-up visual saliency detection algorithm. Different from most previous methods that mainly concentrate on image object, we take both background and foreground into consideration. First, we segment the image into superpixels and collect the superpixels on the image border to be the border set. We eliminate the foreground superpix-els by boundary information from border set to form the background seeds set. Based on the background seeds we can calculate a background-based saliency map. Second, we select fore-ground seeds by segmenting the first-stage saliency map via adaptive threshold. Based on the foreground seeds, we can compute a foreground-based saliency map.Third, the two saliency maps are integrated by the proposed unified function. The unified map can not only highlight the foreground but also restrain the background. Finally, we refine the integrated result to obtain a more smooth and accurate saliency map. We balance the unified saliency using the K-Means and get the pixel-based saliency map using the object-based gaussian map.Moreover, the unified formula also proves to be effective in combining the proposed ap-proach with other models.Experiments on three public available data sets demonstrate that the proposed algorithm performs favorably against the state-of-the-art methods. |
