Real-time Shading: Hardware Shading Effects

In this part of the course we will review some examples of shading algorithms that we might want to implement in a real-time or interactive system. This will help us to identify common approaches for real-time shading systems and to acquire information about feature sets required for this kind of system. The shading algorithms we will look at fall into three categories: realistic materials for local and global illumination, shadow mapping, and finally bump mapping algorithms. 1 Realistic Materials In this section we describe techniques for a variety of different reflection models to the computation of local illumination in hardware-based rendering. Rather than replacing the standard Phong model by another single, fixed model, we seek a method that allows us to utilize a wide variety of different models so that the most appropriate model can be chosen for each application. 1.1 Arbitrary BRDFs for Local Illumination We will first consider the case of local illumination, i.e. light that arrives at objects directly from the light sources. The more complicated case of indirect illumination (i.e. light that bounces around in the environment before hitting the object) will be described in Section 1.3. The fundamental approach for rendering arbitrary materials works as follows. A reflection model in reflection model in computer graphics is typically given in the form of a bidirectional reflectance distribution function (BRDF), which describes the amount of light reflected for each pair of incoming (i.e. light) and outgoing (i.e. viewing) direction. This function can either be represented analytically, in which case it is called a reflection model), or it can be represented in a tabular or sampled form as a four-dimensional array (two dimensions each for the incoming and outgoing direction). The problem with both representations is that they cannot directly be used in hardware rendering: the interesting analytical models are mathematically too complex for hardware implementations, and the tabular form consumes too much memory (a four-dimensional table can easily consume dozens of MB). A different approach has been proposed by Heidrich and Seidel [9]. It turns out that most lighting models in computer graphics can be factored into independent components that only depend on one or two angles. These can then be independently sampled and stored as lowerdimensional tables that consume much less memory. Kautz and McCool [12] described a method for factorizing BRDFs given in tabular form into lower dimensional parts that can be rendered in a similar fashion. As an example for the treatment of analytical models, consider the one by Torrance and Sparrow [29]: