Removing Artifacts Due To Frequency-Domain Processing of Light-Fields

Abstract

In previous works, light-field capture has been analyzed in spatio-angular representation. A light-field camera samples the optical signal within a single photograph by multiplexing the 4D radiance onto the physical 2D surface of the sensor. Besides sampling the light field spatially, methods have been developed for multiplexing the radiance in the frequency domain by optically mixing different spatial and angular frequency components. The mathematical method for recovering the multiplexed spatial and angular information from the frequency representation is very straightforward. However, the results are prone to lots of artifacts due to limitations inherent to frequency-domain processing of images. In this paper, we try understand the characteristics of these artifacts. Furthermore, we study the effect and sources of artifacts that affect the quality of the results and present various methods for the removal of artifacts.