blind motion deblurring from a single image using sparse approximation jian-feng caiy, hui jiz,...
TRANSCRIPT
Blind motion deblurring from a single Blind motion deblurring from a single image using sparse approximationimage using sparse approximation
Jian-Feng Caiy, Hui Jiz, Chaoqiang Liuy and Zuowei ShenzNational University of Singapore, Singapore 117542
Center for Wavelets, Approx. and Info. Proc.y and Department of Mathematicsz
112/04/18 1
報告者:黃智勇
CVPR 2009
OutlineOutlineIntroductionTight framelet system and curvelet systemSparse representation under framelet and
curvelet systemFormulation of our minimizationNumerical algorithm and analysisExperiments
112/04/18 2
IntroductionIntroduction
112/04/18 3
We propose to use framelet system (Ron and Shen et al. [24]) to find the sparse approximation to the image under framelet domain.
We use the curvelet system (Candes and Donoho [8]) to find the sparse approximation to the blur kernel under curvelet domain.
Tight famelet systemTight famelet system
112/04/18 4
Sparse representation under Sparse representation under framelet andframelet and curvelet systemcurvelet system
112/04/18 5
Formulation of our minimizationFormulation of our minimization
112/04/18 6
We denote the image g (or the kernel p) as a vector g (or p). Let “ 。” denote the usual 2D convolution after column concatenation, then we have
Let u = Ag denote the framelet coefficients of the clear image g, and let v = Cp denote the curvelet coefficients of the blur kernel p.
Numerical algorithm and analysisNumerical algorithm and analysis
112/04/18 7
there exist only two difficult problems (14) and (15) of the same type. For such a large-scale minimization problem with up to millions of variables, there exists a very efficient algorithm based on so-called linearized Bregman iteration technique.
Numerical algorithm and analysisNumerical algorithm and analysis
112/04/18 8
ExperimentsExperiments
112/04/18 9
ExperimentsExperiments
112/04/18 10
ExperimentsExperiments
112/04/18 11