Monte Carlo Noise Removal Based on Guided Image Filter and Weighted Local Regression

A novel adaptive rendering method based on guided image filter(GIF)and weighted local regression is proposed to remove Monte Carlo(MC)noise while preserving image details.First,GIF is employed to prefilter features when there are complex motions,textures and geometries.In this case,our method produces impressive results for specific effects such as depth of field and motion blur.Second,the images are reconstructed using weighted local theory,which fits the parametric curves in a local space for bias and variance,respectively.As a result,the impact of noise propagation is reduced greatly.Finally,Stein’s unbiased risk estimator(SURE)is adopted to recognize complex pixels and direct the adaptive sampling process.Experimental results demonstrate that our method outperforms prior algorithms in terms of visual image quality and numerical error.In addition,the new method can handle a wide variety of specific distributed effects.