Matthias B. Hullin
Light transport and optical phenomena
Whether in the synthesis or the analysis (interpretation) of images, we have to use strongly simplifying models for the physical nature of light and its interaction with the real world. My work deals with the development of light transport models that are closer to physics but still tractable in terms of storage and computational effort. By using simple but powerful extensions on top of established approaches, we have introduced techniques that achieved results of unprecedented quality in applications as diverse as 3D scanning, photorealistic rendering (offline and real-time), and display technology. Many unsolved problems will benefit from improved models of light transport.
Forward and backward graphics
My research is dedicated to the development of novel solutions to challenges in computer vision and image processing. In recent research projects, we have demonstrated the potential of analysis-by-synthesis techniques for problems such as the tomographic reconstruction of fluid mixing processes. This emerging family of methods marries a synthesis step (forward model of image formation) to an optimization loop, in order to solve the inverse problem of recovering the model parameters that best explain the given set of measured data. The forward step can be as complex as a full global illumination renderer! The resulting objective functions are often mathematically ill-behaved and therefore require careful selection and adaptation of suitable optimization schemes.
The advent of digital image sensors, together with today's numerical methods and advanced compute hardware, allows us to revisit the history of engineering. A recurring theme in my work is to question established design principles that date back to the pre-computer age. For instance, we recently demonstrated a computational camera that, instead of a complex lens with up to several dozen individual glass elements, uses much simpler optics while still maintaining excellent picture quality. Other recent developments include an affordable and practical system to image light in flight, paving the way for temporal analysis of multi-path illumination. Thanks to current developments, in particular sparse codings, many other applications will benefit as we begin to treat computation as an integral component of the measurement process.