Pradeep Sen, Dept. of Electrical & Computer Engr., University of New Mexico
Recent progress in computer graphics has benefited our society in many ways: from entertainment (e.g. movies and games) to product manufacturing (e.g. virtual prototyping) and medicine (e.g. interactive medical visualization). However, despite these improvements we are still far from true interactive photorealism. In this research, the investigators develop a novel framework for computer graphics that improves the speed and quality of existing algorithms by leveraging ideas from the emerging field of compressed sensing. By taking advantage of the compressibility of real-world signals, the researchers explore new algorithms for image synthesis and acquisition. The broader impact of this work is that the core ideas developed will not only benefit important applications in computer graphics, but could also impact areas such as Magnetic Resonance Imaging (MRI) used for medical applications. On the educational side, the PI integrates Hispanics students into the research by fostering relationships with Latin America.
This research is developing a fundamentally new paradigm for a core area of computer graphics: sampling and reconstruction. Most graphics algorithms (e.g. rendering systems) expend their effort sampling the entire signal, despite the fact the signal will be compressed afterwards (e.g. with a transform-coding compression algorithm such as JPEG). The investigators apply the ideas of compressed sensing in order to take advantage of the sparsity in the transform domain and sample the signal in an efficient manner. This results in a framework that can be used to accelerate rendering algorithms by reconstructing the final image from a sparse set of samples using greedy optimization algorithms. The same framework can also be used to accelerate the acquisition of light transport which is useful for relighting applications. The fundamental science explored through this work will spur new areas of research within the graphics community and in related fields.
