A Comparative Study for Single Image Blind Deblurring
A Comparative Study for Single Image Blind Deblurring Wei-Sheng Lai Jia-Bin Huang Zhe Hu UC Merced UIUC UC Merced Narendra Ahuja Ming-Hsuan Yang UIUC UC Merced
Single Image Blind Deblurring • Algorithms: Fergus et al. 2006 Xu et al. 2013 Shan et al. 2008 Zhong et al. 2013 Cho & Lee 2009 Sun et al. 2013 Krishnan et al. 2011 Michaeli et al. 2014 Whyte et al. 2011 Pan et al. 2014 Hirsch et al. 2011 Perrone et al. 2014 • Datasets: Levin et al. 2009 • Real images: Depth variation Kohler et al. 2012 Camera response functions Sun et al. 2013 Saturation Compression artifacts
Our Goal • Performance evaluation on real-world blurred images – a dataset of real images – large scale comparative study 3
User-Study • 4
From Paired Comparisons to Full Ranking • Cumulative Frequency B-T Model 5
Comparing Real and Synthetic Datasets 6
Comparing Image Quality Metrics Full-reference metrics No-reference metrics 7
Observations • Image priors: sparse priors are more robust than edge prediction methods • Image formations: • Datasets: performance on synthetic datasets does not reflect the performance on real images • Quality metrics: IFC/VIF > PSNR/SSIM; none of noreference metrics are applicable 8
Conclusions • First large scale comparative study on real-world images – quantitatively evaluate the progress of the field – identify potential research directions • Code, datasets and results are available: bit. ly/deblur_study • Poster #22 9
- Slides: 9
