Enforcing Consistency of 3D Scenes with Multiple Objects Using Shape-from-Contours Matthew Grum and Adrian G. Bors Dept. of Computer Science, University of York, York YO10 5GH, UK Abstract. In this paper we present a new approach for modelling scenes with multiple 3D objects from images taken from various viewpoints. Such images are segmented using either supervised or unsupervised al- gorithms. We consider the mean-shift and support vector machines for image segmentation using the colour and texture as features. Back- projections of segmented contours are used to enforce the consistency of the segmented contours with initial estimates of the 3D scene. A study for detecting merged objects in 3D scenes is provided as well. Keywords: 3D scene reconstruction, multi-view images, Shape-from- contours, image segmentation. 1 Introduction Single 3D object reconstruction from several images has attracted considerable research interest using various approaches such as multi-view stereo, shape-from- silhouettes, shape-from-shading, etc. Various 3D object representations are used including voxels [1], radial basis functions (RBF) [2] and meshes [3]. Space carv- ing is a form of multi-view stereo which assigns voxels to a 3D object or carves them away from its volume according to their photoconsistency [1]. Implicit RBFs have been shown to represent well surfaces of 3D objects in [2]. Object reconstruction from its silhouettes was based on the principle of duality between the tangent planes and their corresponding object space [4,5] or by using visual hulls of the object [6]. In [7] disparities between projections of 3D patches from various images, are used to correct the 3D scene modelled using RBFs. The methodology described in this paper aims to robustly enforce the con- sistency of scenes of multiple objects with their corresponding contours seg- mented from images. We consider a certain approximate representation of the 3D scene as the initialization. The surface representation elements, which may be voxels, RBF centers or vertices, are displaced according to their consistency with segmented contours. We consider both supervized and unuspervized seg- mentation for extracting object contours from images from various views. The shape-from-contours approach for correcting 3D scenes of multiple objects by using object contour consistency is described in Section 2. A study of identifying R. Wilson et al. (Eds.): CAIP 2013, Part I, LNCS 8047, pp. 539–547, 2013. c  Springer-Verlag Berlin Heidelberg 2013