Inferring the kernel: multiscale method Input image Loop over scales Variation al Bayes Upsample estimates ulti-scale approach to avoid local minim Initialize 3x3 blur kernel Convert to grayscale Remove gamma correction User selects patch from image
Mar 27, 2015
Inferring the kernel: multiscale method
Input image
Loop over scales
Variational
Bayes
Upsample
estimates
Use multi-scale approach to avoid local minima:
Initialize 3x3 blur kernel
Convert tograyscale
Remove gammacorrection
User selects patch from image
Image ReconstructionInput image
Full resolutionblur estimate
Non-blind deconvolution
(Richardson-Lucy)
Deblurred image
Loop over scales
Variational
Bayes
Upsample
estimates
Initialize 3x3 blur kernel
Convert tograyscale
Remove gammacorrection
User selects patch from image
Results on real images
Submitted by people from their own photo collections
Type of camera unknown
Output does contain artifacts– Increased noise
– Ringing
Compares well to existing methods
Original photograph
Blur kernel Our output
Original photographMatlab’s deconvblind
Original
Our output
Close-up of garland
Matlab’sdeconvblind
A submitted photograph
A small list of the reasons why we didn’t attempt this photograph:
• Most of the features of interest are saturated.
• Different blur kernels for different lights (compare lantern streaks with sky light streaks and different than the water reflection streaks, and the car streaks below bridge and the streaks to left of bridge.)
• The objects reflected by flash are stationary and have no motion blur.
We don’t handle subject motion blur
Failure mode
Original photograph
Matlab’s deconvblind
Photoshop sharpen more
Our output Blur kernel
Original photograph
Our output
Blur kernel
Original photograph
Our output
Blur kernel
Matlab’s deconvblind
Original photograph