Nowadays the task of material reflectance properties reconstruction is applied for design and control of projected product appearance and for preliminary calculation of illumination in projected buildings. For designer it is much easier to obtain material models from reality than create them manually. For preliminary illumination calculation we should obtain physically correct lighting models of real materials. The necessity in such models is obvious. We need them in computer games, films and VR (virtual reality) systems. With the aim of our technology for material reconstruction you can obtain reflectance properties of materials and use them for physically correct and photorealistic rendering.
The main purpose of this project is image-based reconstruction of material reflective properties. In general case we have several images of object and its geometry (obtained by 3d scaner) as an input of a system. As an output we obtain the model of material in a special representation. Our system supports the following material models:
All these models possess different properties and impose different restrictions on input data.
- BRDF (uniform and non-uniform grid)
- Analytical material models (like Phong, Ward, Cook-Torrence etc.)
- Special analytical-grid model
BRDF is a physically correct material model that describes the widest class of non-transparent materials. Due to its physical correctness we can use it for physically correct rendering (for example, for preliminary calculation of illumination). Although BRDF requires a big amount of images for adequate and full reconstruction. We also should solve problems with interpolation and extrapolation of data.
On the other hand, analytical models don't impose big restrictions on the input data. We can obtain material model by one image. The problem is that the class of reconstructed materials becomes too narrow. Such models also can be used for photorealistic rendering (not physically correct). For example, Phong model is not physically correct.
Analytic-grid model is the compromiss between BRDF and analytical models. There are no big restrictions on input (5-6 images is enough), class of represented materials by this model is wider than by analytical one.
The developed system solves several imporant tasks:
The material reconstruction task was also solved for very important case when the object surface is flat or cylindrical (very useful for car paints reconstruction).
- Automatic fitting of polygonal object in virtual scene
- Camera and light calibration
- Building of BRDF point cloud by images
- Building a concrete material model (from mentioned above)
- Real-time rendering of material model (GPU accelerated)