Enhancing Bayesian Estimators for Removing Camera Shake

Simple item page
dc.contributor.authorWang, C.en_US
dc.contributor.authorYue, Y.en_US
dc.contributor.authorDong, F.en_US
dc.contributor.authorTao, Y.en_US
dc.contributor.authorMa, X.en_US
dc.contributor.authorClapworthy, G.en_US
dc.contributor.authorYe, X.en_US
dc.contributor.editorHolly Rushmeier and Oliver Deussenen_US
dc.date.accessioned2015-02-28T16:07:15Z
dc.date.available2015-02-28T16:07:15Z
dc.date.issued2013en_US
dc.description.abstractThe aim of removing camera shake is to estimate a sharp version x from a shaken image y when the blur kernel k is unknown. Recent research on this topic evolved through two paradigms called MAP(k) and MAP(x,k). MAP(k) only solves for k by marginalizing the image prior, while MAP(x,k) recovers both x and k by selecting the mode of the posterior distribution. This paper first systematically analyses the latent limitations of these two estimators through Bayesian analysis. We explain the reason why it is so difficult for image statistics to solve the previously reported MAP(x,k) failure. Then we show that the leading MAP(x,k) methods, which depend on efficient prediction of large step edges, are not robust to natural images due to the diversity of edges. MAP(k), although much more robust to diverse edges, is constrained by two factors: the prior variation over different images, and the ratio between image size and kernel size. To overcome these limitations, we introduce an inter‐scale prior prediction scheme and a principled mechanism for integrating the sharpening filter into MAP(k). Both qualitative results and extensive quantitative comparisons demonstrate that our algorithm outperforms state‐of‐the‐art methods.The aim of removing camera shake is to estimate a sharp version x from a shaken image y when the blur kernel k is unknown. Recent research on this topic evolved through two paradigms called MAP(k) and MAP(x,k). MAP(k) only solves for k by marginalizing the image prior, while MAP(x,k) recovers both x and k by selecting the mode of the posterior distribution. This paper first systematically analyzes the latent limitations of these two estimators through Bayesian analysis. We explain the reason why it is so difficult for image statistics to solve the previously reported MAP(x,k) failure. Then we show that the leading MAP(x,k) methods, which depend on efficient prediction of large step edges, are not robust to natural images due to the diversity of edges. MAP(k), although much more robust to diverse edges, is constrained by two factors: the prior variation over different images, and the ratio between image size and kernel size.en_US
dc.description.number6
dc.description.seriesinformationComputer Graphics Forumen_US
dc.description.volume32
dc.identifier.issn1467-8659en_US
dc.identifier.urihttps://doi.org/10.1111/cgf.12074en_US
dc.publisherThe Eurographics Association and Blackwell Publishing Ltd.en_US
dc.subjectblind deconvolutionen_US
dc.subjectBayesian estimatoren_US
dc.subjectimage deblurringen_US
dc.subjectI.3.7 [Computer Graphics]en_US
dc.subjectImage Processing and Computer Visionen_US
dc.subjectI.4.3 [Enhancement]en_US
dc.subjectSharpening and Deblurringen_US
dc.titleEnhancing Bayesian Estimators for Removing Camera Shakeen_US
Files
Collections
32-Issue 6