Abstract

The generation of photorealistic images is a very complex field inside computer graphics. The difficulty of the problem lies in the fact that to generate these images it is necessary to physically simulate correctly all the possible interactions of light with the medium. The most commonly used techniques are based on stochastic methods, but due to their random behavior, they generate noise in the final image. This noise is visible mainly in those areas that are perceptually soft, and to eliminate it, calculation processes that can be very long are necessary.

In this work, we present a different method based on quadrature techniques whose convergence is better than stochastic methods in areas with soft illumination. This method works in a deterministic way using an adaptive scheme, which allows distributing the work to those areas of increased difficulty (given an estimation of the scene error), reducing the computation time and increasing the accuracy. For this purpose, a data structure optimized for this problem is used to minimize the computation time. Finally, the use of a hybrid method between the stochastic approach and our proposed method, which seeks to take advantage of the benefits of each one while combining them, will be studied. We will integrate our proposed method into the architecture of the Mitsuba renderer and validate it against an established standard technique called Path Tracing implemented in this software.

Bibtex

@article{crespo2018quad,
      author        = "Crespo, Miguel and Mu{\~n}oz, Adolfo",
      title         = "{Renderizado adaptativo mediante técnicas de cuadratura}",
      year          = "2018",
}

Acknowledgements

This project has been developed at the Graphics and Imaging Lab at the University of Zaragoza. I would like to thank all its members, and especially Ibón, for making me feel at home and for all the help provided during the months I have been with them. I would especially like to thank Adolfo Muñoz, my supervisor, for his expertise, support and help during the completion of this work.