Refinement Criteria And Its Application In Computer Graphics

In several domains a refinement criterion is often needed to decide whether to go on or stop sampling a signal. When the sampled values are homogeneous enough, we assume that they represent the signal fairly well and we do not need further refinement, otherwise more samples are required. For this purpose, the design of a criterion which is sensitive to small discrimination is very important and these criteria also take a great part in computer graphics.Realistic image synthesis technology is a significant part in computer graphics. When using this technology to compute and synthesize images, it is must on the basis of a certain illumination model the relatively most important one of which is global illumination. It is based on the physical theory to simulate the transmission of light in virtual environment and can be summed as the addition of the integration the dimension of which is increasing continually or even infinite. So it is a sophisticated problem. Generally speaking, Monte Carlo is the only choice to correctly compute the problem of global illumination in the field of realistic image synthesis. However, the image produced by a Monte Carlo based global illumination algorithm is noisy when not using a large enough number of samples. Adaptive sampling is an attractive means to reduce the image noises and its core is the correct selection of the pixel quality criterion.In this thesis, we investigate the characteristics of Tsallis entropy in the domain of information theory and AG-distance refinement criterion and introduce them into the adaptive sampling for Monte Carlo global illumination. In the use of Tsallis entropy, by utilizing the least-squares design, an optimal entropic index q can be obtained automatically to run adaptive sampling effectively. We use the RMS tool to evaluate the qualities of the synthesized images and a great many experimental results demonstrate the effectiveness of Tsallis entropy and AG-distance refinement criteria. Moreover, we attempt to introduce the Shannon entropy into the mesh simplification and have received an elementary achievement.