Rendering Multi-Layered Translucent Materials

Kurt Berglund - CS348B Final Project Proposal

"Click here for the final writeup! [KurtBerglund/FinalProject]"

Overview

For my project, I plan on extending pbrt to be able to shade multi-layered translucent materials such as leaves, paint, skin, or paper. As a final deliverable, I plan on focusing on rendering leaves. The photos below of plants in the courtyard of my apartment complex are an example of what I hope to render. As we have talked about in class, creating a realistic plant rendering is a very difficult process, so I hope that by implementing some recent papers on subsurface scattering in translucent materials, as well as extending upon them in the context of my project, I can attempt to create a realistic looking plant.

Since the papers I am drawing on for my project are fairly new, it will likely be technically challenging to implement the subsurface shading model and require some major additions to pbrt, but because it is fairly new research, should also provide opportunity to try out and explore new ideas that will also present technical challenges. Due to the added realism these models can produce though, it should come with a nice payoff in the final images. Also, getting the right parameters to represent the leaves should also prove to be a technical challenge. I plan on looking in to how leafs are structured to better approximate their characteristics. The papers referenced below contain some sample parameters which should be helpful as an initial setup, but since the leaves I plan to render are available outside my apartment, I also am hoping to find an easy method to somehow measure their properties, although this is a reach goal.

Here are two images that will serve as models for what my final rendering will look like. The first image is a standard leaf, but has interesting features to model such as the veins. The second plant image contains an interesting pattern within the leaf that could also be fun to render. Based upon my findings with implementing a subsurface scattering model, I'll then probably decide on the final look of the leafs I'm looking to render.

Approach

Step 1: Get Up to Speed on Current Subsurface Scattering Approaches

My first step is to just get up to speed on subsurface scattering methods. The main paper I am drawing upon is "Light Diffusion in Multi-Layered Translucent Materials" by Craig Donner and Henrik Wann Jensen, which describes a shading model for light diffusion in multilayered translucent materials. I also plan on consulting works that they cite and have extended upon, such as that described in "A Practical Model for Subsurface Light Transport" By Henrik Wann Jensen et. al., as well as earlier work that both papers reference.

Step 2: Implement Shading Model for Multilayered Translucent Materials

After learning about the approaches, my next step will be to implement the algorithm in the above paper, or to begin on any extensions I think of while doing step 1. I will most likely just implement the paper at this point though so that I have a baseline for further parts. When doing renderings, I plan on just using a simple leaf model (found either by looking on the web, or generating myself), to use to render. I then plan on extending or tweaking the algorithm until I find something that gives good results.

Step 3: Add Realism to Plant

Once I have a good multi-layer subsurface scattering model, I'll then move on to adding more realism to the plant outside of just the material properties. This could include using a bump map to simulate the veins on the leaf, or by modeling the veins using geometry, or using a texture to simulate the neat appearance of the second image above.

In terms of a time frame, I hope to finish step 1 this weekend, and then move on to recreating the subsurface scattering model this coming week. Then I hope to have a week to explore the subsurface scattering model before adding some more realism to the plant. I hope to be done with the actual project a few days before the presentations to allow as much time as possible for the rendering (since it appears people in the past have rendered their final images over the course of several days).