An Evaluation of Appearance Models for Cloth Rendering
Cloth is a common material which exhibits complex, anisotropic reflection behavior. Because cloth is such an important material class in human-centric environments, many models have attempted to describe its appearance. This thesis aims to clarify the state of the art of cloth rendering by comparing two recent models, by Irawan et al. and by Sadeghi et al., which both describe cloth as woven microcylinders. We fit each of these models to a set of measured fabrics and link the resulting differences to theoretical elements of the models. Through this comparison, we find that Sadeghi's model is able to more closely match the appearance of complex fabrics.
Additionally, we present measurements of 8 fabric samples, which exhibit several different structural characteristics. These samples include woven, knitted, nap and pile fabrics. Sadeghi's model is used to reproduce the appearance of the samples. This both allows us to further evaluate this model, and expands the library of cloth types which Sadeghi's model has been used to describe. We find the model to be largely successful in matching the samples, although we also discover a key limitation inherent to woven cloth models. To address this, we suggest an extension, called the thread direction curve which would further generalize the model.
The measurements in this thesis have been included in a supplementary archive.