Detection and Removal for Impulse Noise in Monte Carlo Global Illumination Rendered Images of Highly Glossy Scenes

Monte Carlo global illumination (MCGI) plays an important role for the rendering and synthesis of photorealistic image. However, the synthetic images using MCGI are always corrupted by heavy noise. Especially, very unpleasant impulse noise appears in MCGI images when 3D environments contain highly glossy surfaces. In this paper, we propose a two-stage filtering technique, by utilizing pixel samples, to remove the impulse noise during MCGI image rendering. Firstly, impulsive pixels are detected using the impulsiveness of the distribution. And then the noise suppression technique is applied to obtain the final result. For a salt-like noise, the outlier samples are identified by the Grubbs test and their energies are re-distributed to the neighbors of the noisy pixel. For a pepper-like noise, the energies of its neighbors are gathered. Experimental results show that the proposed method behaves remarkably well in terms of both quantitative assessment of noise reduction and qualitative evaluation of image quality.