NTIRE 2018 Challenge on Spectral Reconstruction from RGB Images

This paper reviews the first challenge on spectral image reconstruction from RGB images, i.e., the recovery of whole-scene hyperspectral (HS) information from a 3-channel RGB image. The challenge was divided into 2 tracks: the "Clean" track sought HS recovery from noiseless RGB images obtained from a known response function (representing spectrally-calibrated camera) while the "Real World" track challenged participants to recover HS cubes from JPEG-compressed RGB images generated by an unknown response function. To facilitate the challenge, the BGU Hyperspectral Image Database [4] was extended to provide participants with 256 natural HS training images, and 5+10 additional images for validation and testing, respectively. The "Clean" and "Real World" tracks had 73 and 63 registered participants respectively, with 12 teams competing in the final testing phase. Proposed methods and their corresponding results are reported in this review.

1. Introduction

Hyperspectral imaging systems (HISs) record the complete spectral signature reflected from each observable point in a given scene. While HISs have been available since the 1970s [8], recent technological advances have reduced their cost and made them accessible to a growing number of researchers and industrialists. Despite their increasingly lower cost, most HISs still rely on either spatial or spectral scanning (via push-broom or filter-wheel principles) in order to acquire complete hyperspectral (HS) images. This inherent limitation of traditional HISs makes them unsuitable for rapid acquisition, or acquiring scenes which contain moving objects. In addition, most HISs are still too physically large and heavy to fit most portable platforms such as drones, smartphones, and other hand-held devices.