BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Deep Dictionary Learning Approaches for Image Super-Resolution - P
 ier Luigi Dragotti (Imperial College London)
DTSTART:20200305T150000Z
DTEND:20200305T160000Z
UID:TALK132442@talks.cam.ac.uk
CONTACT:Carola-Bibiane Schoenlieb
DESCRIPTION:Single-image  super-resolution  refers  to  the  problem of ob
 taining  a  high-resolution  (HR)  version  of  a  single low-resolution  
 (LR)  image. This problem is  highly  ill-posed since it  is  possible  to
  find  many  high-resolution  images that  can  lead  to  the same low-res
 olution one.\n\nCurrent strategies  to  solve  the  single-image super-res
 olution problem are learning-based and the model that maps the LR image to
  the HR image is learned from external image datasets.\n\nOriginally\, lea
 rning-based approaches were built around the idea that both the LR and HR 
 images admit a sparse representation in proper dictionaries and that the s
 parsity patterns of the two representations can be shared when the design 
 of the two dictionaries is properly coupled. More recently\, deep neural n
 etwork (DNN) architectures have led to state of the art results.\n\nInspir
 ed by the recent success of deep neural networks and the recent effort to 
 develop multi-layer sparse models\, we propose an approach based on deep d
 ictionary learning. The proposed architecture contains several layers of a
 nalysis dictionaries to extract high-level features and one synthesis dict
 ionary which is designed to optimize the reconstruction task. Each analysi
 s dictionary contains two sub-dictionaries: an information preserving anal
 ysis dictionary (IPAD) and a clustering analysis dictionary (CAD). The IPA
 D with its corresponding thresholds passes the key information from the pr
 evious layer\, while the CAD with its properly designed thresholds provide
 s a sparse representation of input data that facilitates discrimination of
  key features.\n\nWe then look at the multi-modal case and use the diction
 ary learning framework as a tool to model dependency across modality\, to 
 dictate the architecture of a deep neural network and to initialize the pa
 rameters of the network. Numerical results show that this approach leads t
 o state-of-the-art results. 
LOCATION:MR 14
END:VEVENT
END:VCALENDAR