The usual patch-wise sparse coding of images has been very successful, but leads to a representation that is not optimal for the image as a whole. In contrast, the recently developed convolutional sparse coding computes a representation for an entire image, but broader use of this model has been hampered by the high computational cost. A new efficient algorithm will be presented, and some applications enabled by this algorithm will be discussed.

Speaker: Brendt Wohlberg

Dr. Brendt Wohlberg received the BSc (Hons) degree in applied mathematics, and the MSc (Applied Science) and PhD degrees in electrical engineering from the University of Cape Town (UCT), South Africa in 1990, 1993 and 1996 respectively. After graduating, he held postdoctoral research positions at UCT and thereafter at Los Alamos National Laboratory (LANL). He has been a technical staff member in Theoretical Division since 2002. His research interests include image restoration and related inverse problems, sparse representations, exemplar-based methods, and machine learning.