Geometrical transformation
This part of the 3D pipeline is very processor-intensive, as all the calculations for the 3D scene are carried out at this stage. Described in simple terms, it comprises the following steps:
KIllumination–The illumination of the scene by different light sources is calculated.
KTransformation–In transformation, the objects are aligned in perspective as seen from the observer's point of view.
KBack face culling–This process computes hidden surfaces resulting from the observation perspective chosen. Any object having an invisible front surface is omitted.
K3D clipping–In this process, each polygon is checked to determine whether it is partially or fully invisible. The invisible faces or parts of objects will be removed.
KScaling on the screen – The above steps are now calculated for three-dimen- sional space using normalized coordinates. The on-screen image coordinates will only now be computed.
Rendering
At this stage, the 3D scene is filled with color shades and textures are applied. Different processes and methods are also applied here.
KTexture mapping – At this stage, the 3D object undergoes a sort of "face lift". The materials and textures are assigned. Different methods are used here to make the textures appear realistic, even when enlarged or reduced. As a first step, the tex- tures are computed:
–Point sampling is the simplest method. A pixel-by-pixel comparison is made between the texture template and the surface to be filled. This method leads to a very coarse representation, especially when enlarged.
–In linear mapping, a new color value is interpolated from the adjacent pixels (or texels) of a texture. This gives better results than point sampling, as the hard boundary between the coarse pixels is blurred.
–The MIP mapping method stores a large number of enlargement stages for the texture. The depth information of a primitive is then used to determine which enlargement stages of the texture will be used in drawing. Normal textures sel- dom contain more than 256 colors.
The first 15 bits of a 16-bit wide color representation are reserved for the colors (5/5/5 > R/G/B). Information concerning the transparency of the texture is car- ried in the alpha channel. The last bit is reserved for this information. Finally, a distinction is made in MIP mapping between bilinear and trilinear filtering. Bilin- ear filtering interpolates between two pixels of two textures, trilinear filtering interpolates between four pixels for each of two textures.
