3-D Geometry Enhanced Superpixels for RGB-D Data

Abstract. This paper introduces a novel 3-D geometry enhanced superpixels for RGB-D data. First, we reconstruct the 3-D geometry of the scene by projecting the depth map into 3-D coordinates. Then, a distance metric for superpixel clustering is constructed using 3-D geometry and color information. Finally, pixels are iteratively clustered into superpixels using the proposed distance metric. The proposed method is able to distinguish objects in similar colors due to the introduced 3-D geometry. The oversegmentation results on RGB-D pairs in the Middlebury datasets demonstrate that our approach shows better performance than other three state-of-the-art superpixel methods. The proposed superpixels are also evaluated in the application of segmentation, and we achieve the best segmentation results compared with three state-of-theart segmentation methods.