 |
Image Deblurring[Click to download a package with the executable and examples] This is an executable for image blind deconvolution. The command is: deblur.exe blurImageFileName outputFileName kernelWidth kernelHeight multiScaleRatio noiseStr deblurStrength kCutRatio weight1 weight2 weight3 weight4 weight5 modelUnfitStr Examples
Parameter Definition
Parameter Tuning
We suggest tuning the parameters
noiseStr and
deblurStrength. In many
cases, you can use the default values for
others. The guideline is as follows.
|
| multiScaleRatio | the width ratio of images in two adjacent scales in our multi-scale implementation. It should be in the range of [0.5, 1]. Increasing this value may improve the result quality with longer computation time. |
| noiseStr | the value is in the range of [0.002, 0.5]. Increasing the value makes the result more smoothed while less sensitive to noise. This value corresponds to $\lambda_1$ in our paper. |
| deblurStrength | the value is in the range of [0.01, 1]. A large value makes the image show more details but risk over-sharpening. This parameter depends on $\tau$ described in our paper. |
| kCutRatio | kernel elements smaller than kCutRatio (which are regarded as noise) are set to zeros. The range of kCutRatio is [0.04, 0.1]. |
| weight1 ~ weight5 | usually with default values. |
System Requirement
Windows XP, 2003 Server, or Vista.
The required memory and the computational time depend on the size of the blur image and the estimated kernel. For the "Picasso" example with size 800x532, the peak memory requirement is about 600 Megabytes and the computational time is approximately 8 minutes on a laptop computer.
Version Information:
Distribution Version: 0.2.1 (Oct. 3rd, 2008) by Qi
Shan and Jiaya Jia ({qshan,
leojia}@cse.cuhk.edu.hk),
CSE department, the Chinese University of Hong Kong. This program is
tested on Windows XP, 2003 Server and Vista, but is still not
guaranteed to be bug-free and work properly with all versions of
Windows. It is for educational use ONLY. If you use this executable
for your academic publication, please acknowledge this work:
@article{hqdeblurring_siggraph2008,
author = {Qi Shan and Jiaya Jia and Aseem Agarwala},
title = {High-quality Motion Deblurring from a Single Image},
journal = {ACM Transactions on Graphics (SIGGRAPH)},
year = {2008},
}