Simulation of Wrinkled Surfaces Revisited

in this thesis, the work on simulation of wrinkled surfaces by Jim Blinn is revisited. approximations for first-order 一阶导数 derivatives were used for evaluation of perturbation of normals but an analysis of the impact of this approximation was never provided. furthermore, Blinn only provides a solution for his simulation on shapes with a known bivariate 二元 surface parametrization which excludes triangular meshes with normals assigned at vertex level. Today, triangular meshes is the de facto 事实上 standard, in the 3D graphics industry, to approximate arbitrary shapes including surfaces which are piecewise smooth. Blinn used the restriction that the 2D coordinate passed to evaluate the surface position is also used as the texture sampling coordinate. To tile or rotate a texture on a patch, modulation of the sampling coordinate is required. This restriction is signi cant, espe-cially for triangular meshes, since texture coordinates assigned to each tri-angle is a modulation of the barycentric coordinate. This thesis remedies all these shortcommings. Finally, recent research on bump map filtering is described and evaluated.

a method known as normal mapping was introduced in recent years and it will be shown here how this approach relates to Jim Blinn’s work. today, normal mapping is almost as common as texture mapping and directly supported in a wide range of graphics tools. however, normal mapping is hard to get right without following a strict set of rules. a wrong implementation can lead to discontinuities in the shading. this thesis is focused on eliminating these visual artifacts. the result is an improved and consistent bump/normal mapping method that results in error and artifact free lighting of bumpy surfaces. the results are verfied and compared visually and shown to outperform leading commerical graphics tools.

1 introduction
in 1978, James Blinn published his paper [Bli78] on simulation of wrinkled surfaces. The aim of his paper was to mimic the high frequency surface irregularities that we see in reality on objects everywhere. his work is particularly focused on parametric surface pathes and as Blinn points out these are smooth surfaces and tend to look artificial becaue of this. since modeling such rich detail using patches would be very expensive and most likely a very daunting 令人生畏的 task, Blinn proposes recording such detail in a bump function/map over the surface patch. hence the method is known as bump mapping. concetually the idea is to displace 取代 every point on the original patch by the assigned bump value along the surface normal (see figure 1). in particular this will perturb the surface normal of the original patch.

Blinn argues that for simulation of wrinkles the displacement height is relatively small so the primary effect of the bumps will be in how the pertubation of the normal influences the reflection of the light and not so much the displaced position itself. thus the initial surface is drawn but the pertubed normal is given, as opposed to the actual surface normal of the patch, as input to the lighting model. by this method, completion of an actual displacement of the surface patch is avoided.

later on, it is pointed out by Cook [Coo84] that though the illusion is convincing, the lack of actual detail is revealed particularly in the sihouette around the patch as seen from the camera. his suggestion is to apply an actual displacement of a dense tessellation of the patch and then draw this instead. of course for high frequency detail this tessellation must be very dense. he refers to this technique as displacement mapping 位移映射.

Blinn bases his work on the assumption that the displacements can be considered arbitrarily small relative to the patch itself. Blinn的工作基于这样的假设:相对于patch本身，位移可以被认为是任意小的。as a result of this an approximation is made which will be covered in section 2.1. though displacement mapping might provide pleasing results for significantly larger displacement values than those of wrinkled surfaces and minor surface irregularities the significance of Blinn’s approximation was not reevaluated by Cook. in section 2.3 we will derive an accurate equation for the evaluation of the perturbed normal and provide a closer study of what the actual requirements are for Blinn’s approximate perturbation to be successful. computer graphics has a tendency to be very forgiving in terms of approximations made so it is quite possible the difference in terms of visual appearance in this case could be negligible. nevertheless, our work in section 2.3 will be a study of the criteria necessary for a mathematical success and not a visual one. we will follow up in section 2.6 with a comparison of the visual differences.

in 1998, an alternative to Blinn’s bump mapping was proposed by J.Cohen[Coh98] to simply replace the perturbed normals entirely by a normal map. 1998年，J。将扰动后的法线完全替换为法线贴图 in section 2.2, we will show how this method relates mathematically to Blinn’s work.