Spectrally-Based Single Image Relighting

Abstract. This paper presents a new spectrally-based single image relighting approach. We first compute the spectral radiance of each stimulus on the monitor when displaying the image and then approximate this radiance to the one produced by each corresponding point in the scene to compute spectral reflectance. Then we use the spectral power distribution (defined as SPD) of a new illuminant to relight the image. The key to our method is the use of the spectral radiance of the stimuli on the monitor to approximate that of the objects in the given image. Since a video can be treated as a series of images, we further extend our algorithm to nighttime effects simulation for daytime videos. Our method can also change the illuminant effects of a video.

• Type: Journal
• Year: 2010
• Where: EDUTAINMENT
• Authors: Xiaoxiong Xing, Weiming Dong, Xiaopeng Zhang, Jean-Claude Paul
• Keywords: EDUTAINMENT 2010, Images, Machine Learning, Spectral Radiance, Spectrally-based Single Image