Deep Specialized Network for Illuminant Estimation

Illuminant estimation to achieve color constancy is an ill-posed problem. Searching the large hypothesis space for an accurate illuminant estimation is hard due to the ambiguities of unknown reflections and local patch appearances. In this work, we propos

  • PDF / 10,735,673 Bytes
  • 17 Pages / 439.37 x 666.142 pts Page_size
  • 68 Downloads / 238 Views

DOWNLOAD

REPORT


2

Department of Information Engineering, The Chinese University of Hong Kong, Hong Kong, China {sw015,ccloy,xtang}@ie.cuhk.edu.hk Shenzhen Institutes of Advanced Technology, CAS, Shenzhen, China

Abstract. Illuminant estimation to achieve color constancy is an illposed problem. Searching the large hypothesis space for an accurate illuminant estimation is hard due to the ambiguities of unknown reflections and local patch appearances. In this work, we propose a novel Deep Specialized Network (DS-Net) that is adaptive to diverse local regions for estimating robust local illuminants. This is achieved through a new convolutional network architecture with two interacting sub-networks, i.e. an hypotheses network (HypNet) and a selection network (SelNet). In particular, HypNet generates multiple illuminant hypotheses that inherently capture different modes of illuminants with its unique two-branch structure. SelNet then adaptively picks for confident estimations from these plausible hypotheses. Extensive experiments on the two largest color constancy benchmark datasets show that the proposed ‘hypothesis selection’ approach is effective to overcome erroneous estimation. Through the synergy of HypNet and SelNet, our approach outperforms state-of-the-art methods such as [1–3].

1

Introduction

The aim of color constancy is to recover the surface color under canonical (usually white) illumination from the observed color. Common computational approaches require estimating the spectral illumination of a scene to correct the extrinsic bias it induces. Illumination estimation can be understood as a process of searching through a hypothesis space to identify the best illuminant. It is often difficult to find a good one since the problem is underdetermined – both the illuminant and surface colors in an observed image are unknown. Finding a good hypothesis of illuminant becomes harder when there are ambiguities caused by complex interactions of extrinsic factors such as surface reflections and different texture appearances of objects. Recent methods [2,4,5] attempt to exploit the exceptional modelling capacity of convolutional network for this problem. We argue that it is still non-trivial to Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46493-0 23) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016  B. Leibe et al. (Eds.): ECCV 2016, Part IV, LNCS 9908, pp. 371–387, 2016. DOI: 10.1007/978-3-319-46493-0 23

372

W. Shi et al.

(a) Input image

(b) The restored image using CNN [2] and the corresponding pixel-wise angular error.

(c) The restored image using DS-Net and the pixel-wise angular error maps for branch-1 and -2 for HypNet. The last image is the angular error map after the selection of SelNet.

Fig. 1. The proposed DS-Net shows superior performance over existing methods in handling regions with different intrinsic properties, thanks to the unique synergy between the hypotheses network (HypNet) and selection network (SelN