10.6.0.8 Limitations

Although this technique does closely approximate bump mapping, there are limitations that impact its accuracy.

Bump Map Sampling

The bump map height function is not continuous, but is sampled into the texture. The resolution of the texture affects how faithfully the bump map is represented. Increasing the size of the bump map texture can improve the sampling of the high frequency height components.

Texture Resolution

The shifting and subtraction steps produce the directional derivative. Since this is a forward differencing technique, the highest frequency component of the bump map increases as the shift is made smaller. As the shift is made smaller, more demands are made of the texture coordinate precision. The shift can become smaller than the texture filtering implementation can handle, leading to noise and aliases effects. A good starting point is to size the shift components so their vector magnitude is a single texel.

Surface Curvature

The tangent coordinate axes are different at each point on a curved surface. This technique approximates this by finding the tangent space transforms at each vertex. Texture mapping interpolates the different shift values from each vertex across the polygon. For polygons with very different vertex normals, this approximation can break down. A solution would be to subdivide the polygons until their vertex normals are parallel to within some error limit.

Maximum Bump Map Slope

The bump map normals used in this technique are good approximations if the bump map slope is small. If there are steep tangents in the bump map, the assumption that the perturbed normal is length one becomes inaccurate, and the highlights appear too bright. This can be corrected by creating a fourth pass, using a modulating texture derived from the original bump map. Each value of the texel is one over the length of the perturbed normal: $1 / \sqrt{{\frac{\partial f}{\partial u}}^2 + {\frac{\partial f}{\partial v}}^2 + 1}$