1. Chapter V Illumination and Shaders

2. What is Illumination?
Illumination or lighting refers to the techniques handling the interaction between light sources and objects.
The lighting models are divided into two categories.
Local illumination considers only direct lighting in the sense that the illumination of a surface depends solely on the properties of the light sources and the surface materials. This has been dominant in real-time graphics.
In the real world, however, every surface receives light indirectly. (Even though a light source is invisible from a particular point of the scene, light can still be transferred to the point through reflections or refractions from other surfaces of the scene.) For indirect lighting, the global illumination (GI) model considers the scene objects as potential lighting sources.
Problems of interactive GI
The cost is often too high to permit interactivity.
The rasterization-based architecture of GPU is more suitable for local illumination.
Current status of GI
Approximate GI instead of pursuing precise GI.
Pre-compute GI, store the result in a texture, and use it at run time.

3. Phong Lighting Model - Diffuse Term
The most popular local illumination method is based on the Phong model. It is composed of diffuse, specular, ambient, and emissive terms.
The diffuse term is based on Lambert's law. Reflections from ideally diffuse surfaces (Lambertian surfaces) are scattered with equal intensity in all directions.
So, the amount of perceived reflection is independent of the view direction, and is just proportional to the amount of incoming light.
Among various light types such as point, area, spot, and directional light sources, let’s take the simplest, the directional light source, where the light vector (l) is constant for a scene.

4. Phong Lighting Model - Diffuse Term (cont’d)
Suppose a white light (1,1,1). If an object lit by the light appears yellow, it means that the object reflects R and G and absorbs B. We can easily implement this kind of filtering through material parameter, i.e., if it is (1,1,0), then (1,1,1)(1,1,0)=(1,1,0) where  is component-wise multiplication.
The diffuse term: R G B R G B
intensity 1
intensity .5 1 1 .5 .5
1x1=1
1x1=1
1x0=0
.5x.5=.25
.5x1=.5
.5x.5=.25
(1,1,0)
(.5,1,.5)

5. Phong Lighting Model - Specular Term
The specular term is used to make a surface look shiny via highlights, and it requires view vector (v) and reflection vector (r) in addition to light vector (l).
Computing the reflection vector

6. Phong Lighting Model - Specular Term (cont’d)
Whereas the diffuse term is view-independent, the specular term is highly view-dependent.
For a perfectly shiny surface, the highlight at p is visible only when  equals 0.
For a surface that is not perfectly shiny, the maximum highlight occurs when  equals 0, but falls off sharply as  increases.
The rapid fall-off of highlights is often approximated by (rv)sh, where sh denotes shininess.
The specular term:
Unlike md, ms is usually a gray-scale value
rather than an RGB color. It enables the
highlight on the surface to end up being
the color of the light source.
cosρ
cos2ρ
cos64ρ

7. Phong Lighting Model - Specular Term (cont’d)

8. Phong Lighting Model – Ambient and Emissive Terms
The ambient light describes the light reflected from the various objects in the scene, i.e., it accounts for indirect lighting.
As the ambient light has bounced around so much in the scene, it arrives at a surface point from all directions, and reflections from the surface point are also scattered with equal intensity in all directions.
The last term of the Phong model is the emissive term me that describes the amount of light emitted by a surface itself.

9. Phong Lighting Model
The Phong model sums the four terms!!

10. Shader
A shader is an executable program running on the GPU, and consists of a set of software instructions.
Shading languages (They are all C-like.)
High Level Shading Language (HLSL) by Microsoft
Cg (C for graphics) by NVIDIA
GLSL (OpenGL Shading Language) by OpenGL ARB

11. Shader (cont’d)
Special data types such as float4 (a four-element vector) and float2x3 (a 2x3-element matrix) are supported.
Two kinds of input data
Varying data: unique to each execution of a shader (e.g., vertex position)
Uniform data: constant for multiple executions of a shader (e.g., world matrix) – specified as global variables or with the keyword uniform
Intrinsic functions such as mul()

12. Shader (cont’d)
All varying inputs must have associated semantic labels such as POSITION and TEXCOORD0. The semantics can be assigned by two methods:
Appending a colon and the semantic label to the parameter declaration.
Defining a structure with semantics assigned to the structure members.
The output data can be described using the keyword out if the data are included in the parameter list. (Otherwise, the return value of a shader is automatically taken as the output.)
All outputs must have associated semantic labels. A shader provides output data to the next stage of the rendering pipeline, and the output semantics are used to specify how the data should be linked to the inputs of the next stage. For example, the output semantic POSITION represents the clip-space position of the transformed vertex that is fed to the rasterizer.

13. Per-vertex Lighting
The vertex shader processes a vertex at a time.

14. Per-vertex Lighting (cont’d)
Recall what the rasterizer does.
The simplest fragment shader for per-vertex lighting

15. Problems in Per-vertex Lighting
Per-vertex lighting followed by “color interpolation” makes the colors at vertices evenly interpolated along the edges and then along the scan lines. So, highlights inside a triangle cannot be caught.
Such a problem of per-vertex lighting can be overcome by employing per-fragment lighting, which is often called a normal interpolation shading.

16. Problems in Per-vertex Lighting (cont’d)
When the camera moves, the scene rendered by per-vertex lighting can have a popping effect.

17. Per-fragment Lighting
The vertex shader computes the clip-space vertex position (It’s required!) and outputs the world-space position and normal of each vertex to the rasterizer.
The fragment shader processes a fragment at a time.

18. Ivan Sutherland
Kendall Station at Boston (MIT)

19. Ivan Sutherland (cont’d)
Turing Award
1966 A. J. Perlis
1969 Marvin Minsky
1971 John McCarthy
1972 E.W. Dijkstra
1975 Herbert A. Simon
1977 John Backus
1980 C. Antony R. Hoare
1983 Ken Thompson
Dennis M. Ritchie
1984 Niklaus Wirth
1988 Ivan Sutherland
1999 F. Brooks R. Rivest, A. Shamir, L. Adleman …
Steven A. Coons Award
1983 Ivan Sutherland
1985 Pierre Bezier
1987 Donald Greenberg
1989 David Evans
1991 Andries van Dam
1993 Ed Catmull
1995 Jose Luis Encarnacao
1997 James Foley
1999 James Blinn
2001 Lance Williams …

20. Ivan Sutherland (cont’d)
Ivan Sutherland's students
Edwin Catmull - founder of Pixar and now president of Walt Disney and Pixar Animation Studios
James H. Clark - founder of Silicon Graphics and Netscape
John Warnock - founder of Adobe
Nolan Bushnell - founder of Atari
Alan Kay - inventor of the Smalltalk language
Henri Gouraud - inventor of Gouraud shading algorithm (per-vertex lighting with Phong model)
Bui Tuong Phong - inventor of Phong shading algorithm (per-fragment lighting with Phong model)

21. Global Illumination – Ray Tracing
When a primary ray is shot and intersects an object, three secondary rays would be spawned: a shadow ray, a reflection ray, and a refraction ray.

22. Global Illumination – Ray Tracing
The refraction ray is computed using the Snell’s law.

23. Global Illumination – Ray Tracing (cont’d)

24. Global Illumination – Radiosity
Radiosity algorithm simulates bounced light between diffuse surfaces. Light hitting a surface is reflected back to the environment. As light bounces around an environment, each surface of the environment works itself as a light source. The radiosity algorithm does not distinguish light sources from the objects to be lit by the light sources.
Radiosity of a surface describes the brightness of the surface, and is defined to be the rate at which light leaves the surface.
All surfaces of the scene are subdivided into patches, and then the form factors among all the patches are computed. The form factor describes the fraction of light that leaves one patch and arrives at another.

25. Global Illumination – Radiosity (cont’d)
In principle, the point-to-point form factor needs to be integrated to define a patch-to-patch form factor. Assuming the radiosities are constant over the extent of a patch, computing the patch-to-patch form factor is reduced to computing a point-to-patch form factor.
The form factor between p and sj is conceptually calculated using the hemisphere placed at p. The form factor is defined to be the area projected on the hemisphere's base divided by the base area.
The hemisphere is often replaced by the hemicube. The per-pixel form factors can be pre-computed. Then, patch sj is projected onto the hemicube. The form factor is the sum of the form factors of the pixels covered by the projection.

26. Global Illumination – Radiosity (cont’d)
Once the form factors are computed, solve the linear system.
The radiosity of each patch is obtained. Then, the irradiance of each patch can also be computed as Bi-Ei.

27. Global Illumination – Radiosity (cont’d)