Fuzziness Based Adaptive Sampling Algorithm

The method based on Monte Carlo integration is the main technology to generatethe high quality, physical realistic image, and the first important key in it is calledpixel sampling, means how to get the proper positions in every pixel of the imagescene to compute the global illumination effect and then give the generated image.This paper give a new sampling algorithm, fuzziness based adaptive sampling, toovercome the usual more noise points in image generated by random samplingalgorithm. The new algorithm improves the sampling strategy based on path tracingglobal illumination algorithm. It treat all the illumination computation results in somepixel as a fuzzy set, compute the pixel computation quality through the fuzziness ofthe fuzzy set and then decide whether add samples or not to reduce the noise of theimage in a fixed computation cost.Fuzziness based adaptive sampling algorithm rely on the fact that for a imagesome pixels converge faster than the others. This means that they only need lesssamples to get precise result than the others. In usual way, all the pixel take a uniformnumber of pixels, consequently when the total number of samples is small the imagehas more random noise, and the number is big lead to computing waste. The newalgorithm take the fuzziness as the sampling quality, and only increase the samplenumber in poor quality pixel, so that it can get more reasonable distribution ofsamples in image when the total computation is fixes.