SuperCut: Superpixel Based Foreground Extraction With Loose Bounding Boxes in One Cutting

Abstract

Interactive image segmentation that uses a bounding box containing the foreground has gained great popularity because of its convenience. However, its performance is often degraded when the bounding box is not tight enough or covers large background regions. To solve this problem, this letter proposes a novel segmentation algorithm called “SuperCut”. This algorithm provides robust segmentation in one cut even with loose bounding boxes. In the proposed SuperCut, superpixels of the whole image are first computed, and then a background similarity map for the superpixels inside the bounding box is created based on Haar-Wavelet feature and pixel intensities. Finally, a max-flow cut is made based on the map. The background similarity map reduces the influence from possible background regions inside the bounding box, which leads to the efficient segmentation of SuperCut. To evaluate the effectiveness of the proposed SuperCut, we compared it with some recent published interactive image segmentation techniques. The experimental results show that the proposed SuperCut outperforms other compared methods especially when using loose bounding boxes.