Research And Implementation Of Filter Importance Sampling Algorithm

Photo-realistic image rendering based on global illumination model is one active research area of computer graphics. Image sampling and reconstruction by samples'information are two important aspects of rendering. There are many excellent and practicable algorithms proposed to avoid the aliasing of image and get desirable visual effect. However, they are usually applied to one stage of rendering.Though doing the deep research on the filter sampling algorithm, this paper chooses Gaussian filter to be combined with the algorithm and propose the Gaussian importance sampling algorithm. The new algorithm makes the relation among three stages of image rendering, which are illumination sampling, pixel sampling and the reconstruction, and considers them as a whole in order to have the optimal effect. Construct Gaussian CDF mapping table according to the information in scene files, and translate initial samples generated from uniformly distributed random variables to follow Gaussian function importance sampling. Choose the samples located in the kernel pixel of Gaussian filter as those finally being used to calculate the illumination, and make use of jittering technique to avoid image aliasing. In theory, use the importance sampling of Monte Carlo integration to calculate the final value of pixel, so in reconstruction we use the box filter covering single pixel. Implement the algorithm in the PBRT rendering system as a new sampler plug-in.Design and complete the comparing experiments. On the basis of the experimental results and data, analyze how the changes of the Gaussian filter parameters will make an impact on RMS value and consuming time of the algorithm, and the reason why the algorithm performs poorly when compared with the random sampling combined with Mitchell filter in the comparing experiments, and then figure out the problem and provide make improving suggestions.