Research On Human Skin Rendering Techniques With Adjustable Optical Parameters

Human skin rendering techniques play important roles in the fields of movie making, video games and medical science. With growing need of rendering photorealistic human skin or even virtual characters, skin rendering has gradually become an important subject in computer graphics. It needs too much computation power to fully simulate light transport in skin, thus reducing the amount of computation, at minimal cost of photorealism, is an important topic in skin rendering. Studying how parameters affect skin appearance not only helps render more realistic skin, but also means a lot in other related fields.This thesis focuses on the topic of human skin rendering techniques with adjustable parameters. The objective is to implement a skin rendering engine that respects biological parameters of human skin, and to study how skin parameters affect its appearance. To achieve the objective, three aspects are studied in the thesis, making up a complete solution for both offline rendering and real-time rendering.First, the Multipole diffusion model is not well suited for materials thinner than 2 mean free paths, which includes skin we are going to study. To address this issue, the thesis proposes a simple-to-implement extension that does not affect performance. The extension solves the issue by separating incoming light into two parts, one for direct transmitting light and one for diffused light. Through comparison with Monte-Carlo methods, the accuracy of the extension method is validated for materials thinner than 2 mean free paths.After that, the thesis focuses on applying the extension method to skin rendering. Through investigation of skin structure from microscale to mesoscale, a skin model suitable for parameter adjusting is built. As for rendering, some improvements for surface point generation and texture usage, inspired by real-time techniques, are proposed. Using the skin model, appearance of skin with different parameters are compared. Caucasian skin, Asian skin and African skin are rendered, which highlights the applicability of the model.The major shortcoming of offline rendering is its speed. In order to get real-time responses while adjusting skin parameters, a “pre-compute and interpolation” method is proposed in conjunction with real-time rendering techniques. Using the pre-compute method, real-time adjusting of skin parameters is achieved. The accuracy and efficiency of the method is proven by comparisons with offline rendering results, performance tests and a video recording.At the end of the thesis the limitations of the research are analyzed. Prospect of future work and potential applications are discussed.Compared with previous research, the innovations of the thesis are as follows:(1) An easy-to-implement extension to the Multipole diffusion model is proposed;(2) Analyzes in detail how skin appearance, especially the color bleeding effect, varies with skin parameters.(3) Achieves real-time adjusting of skin parameters.