Image-based path radiance estimation guided by physical model

This paper presents an image-based method for estimating path radiance by utilizing spectral characteristics of path radiance based on a physical model. Though image-based methods can estimate path radiance without using atmospheric properties, their accuracy is highly dependent on what is captured in a scene and hence inefficient in some cases. The proposed method corrects a path radiance spectrum estimated by the dark object subtraction (DOS) method so that the spectrum meets general spectral characteristics of path radiance. A simplified physical model of path radiance is introduced for this purpose. Experimental results show that the proposed method estimates path radiance more accurately than the DOS method in scenes including vegetation in particular. It is also shown that atmospheric correction incorporating the proposed method reduces correction errors by more than 60 percent on average in comparison with major atmospheric correction algorithms: percentage linear contrast stretch (PLCS) and quick atmospheric correction (QUAC).