Texturing explained

Texturing

Texturing is the application of a 2D image to a 3D object that allows for simulating detail. The simulated detail compensates for the lack of highly complex geometry. For example, a realistic tree would require millions of triangles, with each one simulating different parts of the bark and each leaf. By wrapping a texture around a less complex model it is possible to make the tree appear to have considerably more complexity than it actually has. Though imperfect, this is substantially faster than calculating the geometry that would be necessary to display the full detail.

An important part of texturing is mip-mapping, which offers considerable benefit in reducing texture aliasing (artifacts associated with insufficient scene sampling - anti-aliasing will be covered more in part 2 of this article). Textures are typically either squares or rectangles. With that in mind, when they are mapped over long distances, the shape of the texture will not fit the object (as when objects stretch over distances they become narrower with distance, simulating depth). This effect introduces aliasing in the form of texture swimming. Mip-mapping helps to alleviate this issue by resizing the original texture to a variety of smaller sizes. By doing this, the correctly sized texture can be placed over the object at the appropriate distance. As the object moves in or out, the version will be replaced with the new version of the texture. This significantly reduces texture aliasing artifacts. Banding effects that exist when bilinear filtering is used show the separation of mip-map levels. These two images demonstrate the difference when mip-mapping is and is not being used. The screenshot on the left does not have mip-mapping, while the one on the right does.

No mip-mapping

Mip-mapping enabled

On modern graphics cards there are typically 2-4 pixels written simultaneously, depending on the number of pipelines on the graphics chip. Each pixel has a dedicated pipeline, which is designed specifically for rendering. Assuming that there is a pixel going up the pipeline, the u, v coordinates are taken for the pixel and are used to sample the texture in the correct location. Now assuming that bilinear filtering is being used, four samples are taken for the pixel out of the sample position on the texture. Each sample is a pixel, or color value from a texture (typically referred to as a texel). These four samples are blended together to form the color of the rendered pixel.

Bilinear filtering

Trilinear filtering is a more advanced form of bilinear. This texture-filtering mode can be thought of as being intelligent, as it is aware of mip-mapping. One of the common problems with mip-mapping is found when a single object uses multiple mip-map levels at any give time. This use of multiple mip-map levels causes banding lines between each mip level. Trilinear, just like bilinear, takes four samples from a texture and blends them for a given pixel, but it also takes the four equivalent samples from the next mip-map level and blends those as well. This removes the banding normally associated with mip-mapping blending the mip levels together. Additionally, with more samples being taken, trilinear removes other aliasing artifacts associated with general texture sampling.

Trilinear filtering