Shading-Aware Multi-view Stereo

We present a novel multi-view reconstruction approach that effectively combines stereo and shape-from-shading energies into a single optimization scheme. Our method uses image gradients to transition between stereo-matching (which is more accurate at larg

  • PDF / 8,722,250 Bytes
  • 17 Pages / 439.37 x 666.142 pts Page_size
  • 38 Downloads / 190 Views

DOWNLOAD

REPORT


TU Darmstadt, Darmstadt, Germany Adobe Research, San Francisco, USA

Abstract. We present a novel multi-view reconstruction approach that effectively combines stereo and shape-from-shading energies into a single optimization scheme. Our method uses image gradients to transition between stereo-matching (which is more accurate at large gradients) and Lambertian shape-from-shading (which is more robust in flat regions). In addition, we show that our formulation is invariant to spatially varying albedo without explicitly modeling it. We show that the resulting energy function can be optimized efficiently using a smooth surface representation based on bicubic patches, and demonstrate that this algorithm outperforms both previous multi-view stereo algorithms and shading based refinement approaches on a number of datasets.

1

Introduction

High-quality digitization of real world objects has been of great interest in recent years. The demand for effective and accurate digitization methods is increasing constantly to support applications such as 3D printing and visual effects. Passive reconstruction methods such as multi-view stereo [1] are able to achieve high quality results. However, stereo methods typically operate on image patches and/or use surface regularization in order to be robust to noise. As a result, they often cannot recover fine-scale surface details accurately. These details are often captured by shading variations, and recent work has focused on shadingbased refinement of the geometry obtained from multi-view stereo (or in some cases using depth sensors or template models). Starting from the work of Wu et al. [2] that can only be used for objects with constant albedo, algorithms have evolved to operate on implicit surfaces [3] and real time settings [4]. All these methods treat the coarse input geometry as a fixed ground truth estimate of the shape and use it to regularize their optimization. Consequently, uncertainties in the inital reconstruction method are discarded and cannot be resolved reliably. Another challenge for shading-based refinement techniques is that observed image intensities combine shading and surface albedo. Inferring fine-scale detail Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46487-9 29) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016  B. Leibe et al. (Eds.): ECCV 2016, Part III, LNCS 9907, pp. 469–485, 2016. DOI: 10.1007/978-3-319-46487-9 29

470

F. Langguth et al.

from shading thus requires reasoning about surface albedo. This significantly increases the number of variables in the optimization. Most current techniques either assume constant albedo or apply strong regularization on the albedo, which can often fail on real-world surfaces. In contrast to previous work, we propose a new multi-view surface reconstruction approach that combines stereo and shading-based data terms into a single optimization scheme. At the heart of our algorithm is the observation that stereo-matchi