Domain Specific Data-driven Modeling of Materials

Abstract

Materials can nowadays be measured and rendered quite realistically by using the bidirectional texture function (BTF). However editing of BTFs during the design process of materials is challenging. Utilizing information about the specific type of a material at hand, one can derive domain specific model parameters from a given BTF by solving an inverse problem. For textiles e.g. this could be a specific description for a weaving pattern which can be intuitively edited by cloth designers. Reflection properties are associated to certain structures of the material (e.g. yarns) using image segmentation techniques to allow semantic editing operations like changing the color of a certain yarn. The acquired information about the structure of the material may be further exploited to improve the efficiency of physically based rendering techniques. Our aim is to be able to both resynthesize the image efficiently given the model and predict the appearance of changes to the model.

Ongoing Projects

logo=stoff.png 
In this project we work on the analysis, synthesis and resynthesis of optical material properties of cloth. By estimating domain specific parameters like the weaving pattern and yarn reflection properties from images we obtain a cloth model which can both be visually resynthesized and intuitively edited. We develop new techniques in the context of physically based rendering and image analysis of cloth.