SuperSelect
Christoph Korinke, Tim Claudius Stratmann, Tim Laue, Susanne Boll · 2016
We present the concept of an interactive image segmentation method, which allows a fast and precise extraction of foreground objects from natural images. The method is especially suited for mobile, touch-based devices. The approach combines automatic image segmentation with interactive refinement. In a first step, the user extracts an object using the first stage of the GrabCut algorithm. Any errors in the resulting segmentation can then be corrected via a novel correction step in which the underlying image is over-segmented into superpixels using the SLIC algorithm. The user can select individual superpixels in order to adjust the image segmentation. This image segmentation method is implemented as a full working demo on a Microsoft Surface Pro 3.