Foreground Extraction Algorithm Using Depth Information for Image Segmentation

Abstract

Image segmentation is one of the most important topics in the field of computer vision. So lots of approaches for image segmentation have been proposed, and interactive methods based on energy minimization such as Grab Cut, etc have shown successful results. It, however, is not easy to automate the full process for segmentation because almost all of interactive methods require considerable user interaction. So if additional information is provided to users in order to guide them effectively, we can reduce interaction with them. In this paper we propose an efficient foreground extraction algorithm, which makes use of depth information from RGB-D sensors like Microsoft Kinect and offers users guidance for foreground extraction. Our approach can be applied as a pre-processing for interactive and energy-minimization-based segmentation approaches. Our proposed method is able to segment the foreground from images and give hints to reduce interaction with users. In our method, we make use of the characteristics of depth information captured by RGB-D sensors and describe them using information from structure tensor. And in our experiments we show that for real world images the proposed method separates foreground from background sufficiently well.