Adaptive Sampling Based On Tsallis Theory

Realistic Graphics is an important part of Computer Graphics, and with the extensive application of the realistic image synthesis technology, more creative algorithm is required. The method based on Monte Carlo integration is the main technology to generate the realistic images with high quality. In the process of rendering, discrete pixels are needed to present the continuous image, so in essence, rendering problem is sampling problem. With the beginning of aliasing and anti-aliasing, the thesis describes the adaptive sampling algorithm. Adaptive sampling puts more samples on the complex areas and the less on the simple areas of the image not the same on the whole image because that the calculations of Monte Carlo global illumination of all the pixels are different. Some pixels converging faster only need fewer samples to get accurate values while others need more.After reviewing the classical adaptive sampling algorithms based on the Variance, Contrast, Shannon and f-Divergence, the thesis present Tsallis theory and put forward the adaptive sampling algorithms based on Tsallis Entropy and Tsallis Divergence. The method of least square is introduced into the adaptive sampling using Tsallis Entropy to automatically choose the appropriate entropy indices according to the different scenes, and the noise of the images rendered by the algorithm are less than the counterparts rendered by other classical algorithms, especially on the details and complex areas and the RMS values are smaller; while in the adaptive sampling based on Tsallis Divergence, by way of experiments, several different divergence indices are chosen to verify that the images and RMS values are the improvement to the other classical algorithms.