| Reflectance plays an important role in computer graphics. It describes the appearance of an object with two directional parameters. Reflectance is critical, because it determines the appearance of the object to be synthesized. Reflectance can be determined either by an analytical model, or by evaluating a fit to a measured reflectance data set. In general, analytical models are complex and computationally expensive to evaluate and it is difficult to control the parameters of the model to obtain a desired appearance. A popular method of fitting data is by using a basis function expansion. However, this method requires many basis functions to represent the strongly-peaked data and the result is computationally expensive.; We propose a method to overcome this problem by using a modified N-dimensional multilevel B-spline approximation. Our method fits various reflectance data very well. Multilevel architecture makes it possible to control the accuracy of the fit. A higher level fit uses a denser control mesh and fits more accurately. In addition, the resulting fit is very smooth and efficient to evaluate. The time complexity of evaluation is a constant regardless of the fit level. A higher level fit requires more storage than a lower level fit. The storage might be a problem on memory intensive applications. To overcome this, we represent a data set with two fits, a diffuse fit and a specular fit and we can successfully compress the storage for finer fit without losing major performance from the original method. In addition, by utilizing minimal perfect hashing, we can retrieve the value of each control point efficiently from compressed table. |
