12:00 Removing Camera Shake via Weighted Fourier Burst Accumulation. Guillermo Sapiro, Duke University ECE

Numerous recent approaches attempt to remove image blur due to camera shake, either with one or multiple input images, by explicitly solving an inverse and inherently ill-posed deconvolution problem. If the photographer takes a burst of images, a modality available in virtually all modern digital cameras, we show that it is possible to combine them to get a clean sharp version. This is done without explicitly solving any blur estimation and subsequent inverse problem.

The proposed algorithm is strikingly simple: it performs a weighted average in the Fourier domain, with weights depending on the Fourier spectrum magnitude. The method can be seen as a generalization of the align and average procedure, with a weighted average, motivated by hand shake physiology and theoretically supported, taking place in the Fourier domain. The method's rationale is that camera shake has a random nature and therefore each image in the burst is generally blurred differently.

Experiments with real camera data, and extensive comparisons, show that the proposed Fourier Burst Accumulation (FBA) algorithm achieves state-of-the-art results an order of magnitude faster, with simplicity for on-board implementation on camera phones.

Finally, we also present experiments in real high dynamic range (HDR) scenes, showing how the method can be straightforwardly extended to HDR photography. Video debluring is also possible as we demonstrate.

Joint work with Mauricio Delbracio.

13:30 Learning Low-rank Transformations: Algorithms and Applications.
Guillermo Sapiro, Duke University ECE

We describe a very simple framework to learn a linear transform to simplify data and open the door to significant advances in classification. The proposed method, based on concepts of nuclear norm and low-rank representations, is democratic, in the sense that can be used by any algorithm, such as random forests, deep networks, and subspace methods.

Both theoretical foundations and applications are described. Among the examples, we show state-of-the-art face recognition with orders of magnitude less training data and time; basically a few minutes on a desktop computer instead of weeks on clusters is in the current literature.

Joint work with Qian Qiu.