Recording and playback of camera shake:
benchmarking blind deconvolution with a real-world database

Rolf Köhler

Michael Hirsch

Betty Mohler

Bernhard Schölkopf

Stefan Harmeling
Proc. IEEE European Conference on Computer Vision (ECCV) 2012


NOTE:
1) THIS WEB PAGE LOADS RATHER SLOW.
2) COMPRESSED .JPG IMAGES OF THE DEBLURRED IMAGES WERE USED ON THIS WEB PAGE
THE UNCOMPRESSED .PNG FILES, WHICH WERE USED IN THE BENCHMARK CAN BE DOWNLOADED BELOW .

Abstract
Motion blur due to camera shake is one of the predominant sources of degradation in handheld photography. Single image blind deconvolution (BD) or motion deblurring aims at restoring a sharp latent image from the blurred recorded picture without knowing the camera motion that took place during the exposure. BD is a long-standing problem, but has attracted much attention recently, cumulating in several algorithms able to restore photos degraded by real camera motion in high quality. In this paper, we present a benchmark dataset for motion deblurring that allows quantitative performance evaluation and comparison of recent approaches featuring non-uniform blur models. To this end, we record and analyse real camera motion, which is played back on a robot platform such that we can record a sequence of sharp images sampling the six dimensional camera motion trajectory. The goal of deblurring is to recover one of these sharp images, and our dataset contains all information to assess how closely various algorithms approximate that goal. In a comprehensive comparison, we evaluate state-of-the-art single image BD algorithms incorporating uniform and non-uniform blur models.



Downloads

Paper (pdf)Poster (pdf)


Download Deblurred Images and Kernels (uncompressed .png files)

Cho (zip) Fergus (zip) Hirsch (zip) Krishnan (zip) Shan (zip) Whyte (zip) Xu (zip)


Download Complete 6D Trajectories used in the benchmark

6D Trajectories Kernels computed from the 6D Trajectories using this matlab script or the same script written in phython . •

We additionally recorded camera movements at a frame rate of 120 fps. Those trajectories were not used in the benchmark paper, download.

Instructions for authors of deblurring algorithms, who want to benchmark their algorithm.

1. All blurry and deblurred images

Blur Kernel
1 2 3 4 5 6 7 8 9 10 11 12
Image1 (church) 1,1 1,2 1,3 1,4 1,5 1,6 1,7 1,8 1,9 1,10 1,11 1,12
Image2 (clock) 2,1 2,2 2,3 2,4 2,5 2,6 2,7 2,8 2,9 2,10 2,11 2,12
Image3 (backyard) 3,1 3,2 3,3 3,4 3,5 3,6 3,7 3,8 3,9 3,10 3,11 3,12
Image4 (roof) 4,1 4,2 4,3 4,4 4,5 4,6 4,7 4,8 4,9 4,10 4,11 4,12



2. All blur kernels with the according NU values (sec. 5.1)

Visualisation of the 12 blur kernels used for creating the benchmark dataset.

1 2 3 4 5 6 7 8 9 10 11 12



3. 6D vs. 3D approximation by Whyte et al. and Gupta et al. (sec. 5.2)

40 camera shakes and their 3D approximation using 2 different (by Whyte et al. and Gupta et al.) approaches. The true point grid of the 6D trajectorie (left blur kernel), the approximation by Whyte et al. (middle blur kernel) and the approximation by Gupta et al. (right blur kernel) were shifted and overlapped into one image.







ALL 48 IMAGES ARE SHOWN BELOW. PLEASE MOVE YOUR MOUSE OVER THE AUTHOR'S NAMES TO SEE THEIR DEBLURRING RESULTS.




Image 1, Blur Kernel 1 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 27.577





Image 1, Blur Kernel 2 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 30.529





Image 1, Blur Kernel 3 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 35.034





Image 1, Blur Kernel 4 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 30.234





Image 1, Blur Kernel 5 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 26.643





Image 1, Blur Kernel 6 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 27.617





Image 1, Blur Kernel 7 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 27.222





Image 1, Blur Kernel 8 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 22.120





Image 1, Blur Kernel 9 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 22.541





Image 1, Blur Kernel 10 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 23.248





Image 1, Blur Kernel 11 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 25.582





Image 1, Blur Kernel 12 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 27.059





Image 2, Blur Kernel 1 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 21.611





Image 2, Blur Kernel 2 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 24.692





Image 2, Blur Kernel 3 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 28.996





Image 2, Blur Kernel 4 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 25.148





Image 2, Blur Kernel 5 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 20.858





Image 2, Blur Kernel 6 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 21.889





Image 2, Blur Kernel 7 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 21.907





Image 2, Blur Kernel 8 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 16.438





Image 2, Blur Kernel 9 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 17.023





Image 2, Blur Kernel 10 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 17.269





Image 2, Blur Kernel 11 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 19.578





Image 2, Blur Kernel 12 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 20.249





Image 3, Blur Kernel 1 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 27.530





Image 3, Blur Kernel 2 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 30.435





Image 3, Blur Kernel 3 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 35.483





Image 3, Blur Kernel 4 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 30.928





Image 3, Blur Kernel 5 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 34.807





Image 3, Blur Kernel 6 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 28.295





Image 3, Blur Kernel 7 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 28.199





Image 3, Blur Kernel 8 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 22.386





Image 3, Blur Kernel 9 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 21.716





Image 3, Blur Kernel 10 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 23.405





Image 3, Blur Kernel 11 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 26.147





Image 3, Blur Kernel 12 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 26.282





Image 4, Blur Kernel 1 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 22.828





Image 4, Blur Kernel 2 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 25.732





Image 4, Blur Kernel 3 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 29.834





Image 4, Blur Kernel 4 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 26.061





Image 4, Blur Kernel 5 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 34.587





Image 4, Blur Kernel 6 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 23.083





Image 4, Blur Kernel 7 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 22.423





Image 4, Blur Kernel 8 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 17.270





Image 4, Blur Kernel 9 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 17.990





Image 4, Blur Kernel 10 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 18.077





Image 4, Blur Kernel 11 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 20.274





Image 4, Blur Kernel 12 Back to top

Original Blurry Cho Xu Shan Fergus Krishnan Whyte(lsq) Whyte(Kri) Hirsch
BLURRY
PSNR = 21.818