Scene Segmentation Driven by Deep Learning and Surface Fitting

摘要: This paper proposes a joint color and depth segmentation scheme exploiting together geometrical clues learning stage. The approach starts from an initial over-segmentation based on spectral clustering. input data is also fed to Convolutional Neural Network (CNN) thus producing per-pixel descriptor vector for each scene sample. An iterative merging procedure then used recombine the segments into regions corresponding various objects surfaces. proposed algorithm by considering all adjacent computing similarity metric according CNN features. couples of with higher are considered merging. Finally uses NURBS surface fitting in order understand if selected correspond single surface. comparison state-of-the-art methods shows how method provides accurate reliable segmentation.