Matlab code for “Non-uniform Deblurring for Shaken Images”

This package contains code to perform blind deblurring of non-uniform / spatially-varying blur caused by camera shake, using the model described in (1), applied within the algorithm described in (2) and (3). Please cite these three papers if using this code in an academic publication.

1 Installing

Compile the mex files. In Matlab, cd to the code directory, and run mex_all
Add the code directory to your Matlab path

1.1 Compatibility

The code currently does not work on Windows, I’m not sure how much modification is necessary for this. I would be very interested to hear if you successfully compile the mex files and run the code on Windows
Multi-threaded version requires libpthread to be available (tested on 64-bit Linux Ubuntu 9.10 with Matlab R2007b & R2009a, and 32-bit Mac OS X 10.5 with Matlab R2006b)
- For Windows, you might try pthreads-win32, although I have no experience with this myself. If you have any success with this, please do let me know

2 Running

In Matlab, cd to results directory
Ensure that the code directory is on the Matlab path
Run one of the scripts called deblur_...
To create scripts for your own images, the existing scripts should provide a good start
- A basic deblur_... script has a few components:
  1. Set CONFIG_FNAME for the name of the directory where the outputs will go
  2. Set NON_UNIFORM = 1 / 0 for non-uniform / uniform blur model
  3. Run an image_settings_... script, which contains per-image settings
  4. Run the default_config script, which sets a lot of parameters to their default values
  5. Run the deblur script, which runs the actual algorithm
Note that the algorithm can take a long time to run with the non-uniform blur model, e.g. the included pantheon' example takes 3 1/2 hours on my workstation, withNUM_THREADS = 4` (see below)

2.1 Differences with the code from Fergus et al.

The code is not compatible with the original codes from Miskin & MacKay or Fergus et al., as many of the functions have been restructured and take different arguments. However, you should get similar, if not identical, results to that of Fergus et al., when using the uniform blur model

2.2 A few significant options

Please see the original readme from Fergus et al.’s code for an explanation of more options, and more information on the algorithm in general.

PRESCALE: Factor to downsample original blurry image.
BLUR_KERNEL_SIZE: Size of blur in blurry image after downsampling by PRESCALE.
blur_{x,y,z}_lims = [min, max]: Maximum extent of blur due to contributions of $θ_{X}$ , $θ_{Y}$ , and $θ_{Z}$ . Usually derived from BLUR_KERNEL_SIZE.
focal_length_35mm: Camera’s focal length in 35mm equivalent. This is sometimes available directly in the EXIF tags of an image. If not, the actual focal length should be in the EXIF tags. With this, you will need to know the size of the camera’s sensor in mm (you might try looking on http://www.dpreview.com/reviews/specs.asp). Then,
focal_length_35mm = focal_length / sensor_width_in_mm * 36.
AXIS = [xmin, xmax, ymin, ymax]: Region of (downsampled) blurry image to use for kernel estimation.
FIRST_INIT_MODE_BLUR = 'xbar' / 'ybar' / 'zbar': Initialisation of kernel, depending on approximate shape of blur. One of 3 axes: xbar for approximately vertical blur, ybar for horizontal, or zbar for in-plane rotation.
DISPLAY_EACH_ITERATION = true / false: After each iteration, show the latent image and kernel, and plot the value of the cost function over all iterations.
SAVE_EACH_ITERATION = true / false: Save images of the latent image and kernel after each iteration. Note that this option will cause hundreds of (small) image files to be saved in the results directory.
PLOT_LINESEARCH = true / false: At each iteration, show value of cost function at different points along search direction.
NUM_THREADS: On systems where libpthread is available, the non-uniform blur model can be run multi-threaded for speed. On other systems, NUM_THREADS must be set to 1.

3 Acknowledgements

We would like to thank Rob Fergus for making his code available online (http://cs.nyu.edu/~fergus/research/deblur.html) and for kindly allowing us to release our modified version of his code, as well as James Miskin and David MacKay for also making their original code available online (http://www.inference.phy.cam.ac.uk/jwm1003/).

4 License

The code in this package is based on the code provided by Fergus et al., and as such several parts are subject to their own license. For functions originally distributed by Fergus et al., please see the original readme from Fergus et al.’s code for details. These functions are marked with the header:

Author: Rob Fergus (or other)
Version: 1.0, distribution code.
Project: Removing Camera Shake from a Single Image, SIGGRAPH 2006 paper
Copyright 2006, Massachusetts Institute of Technology

For other functions marked with the header:

Author:     Oliver Whyte <oliver.whyte@ens.fr>
Date:       August 2010
Copyright:  2010, Oliver Whyte
Reference:  O. Whyte, J. Sivic, A. Zisserman, and J. Ponce.
   ``Non-uniform Deblurring for Shaken Images''. In Proc. CVPR, 2010.
URL:        http://www.di.ens.fr/~whyte/deblurring/

the following license applies:

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

5 Bugs

Please send bug reports to oliver.whyte@ens.fr

Bibliography

[1] O. Whyte, J. Sivic, A. Zisserman, and J. Ponce. “Non-uniform Deblurring for Shaken Images”. In Proc. CVPR, 2010.

[2] R. Fergus, B. Singh, A. Hertzmann, S. T. Roweis, and W. T. Freeman. “Removing camera shake from a single photograph”. ACM Trans. Graphics (Proc. SIGGRAPH 2006), 2006.

[3] J. W. Miskin and D. J. C. MacKay. “Ensemble Learning for Blind Image Separation and Deconvolution”. In Advances in Independent Component Analysis, 2000.