 |
Technical Reports |  |
Work Products |  |
Research Abstracts |  |
Historical Collections |  |
|
Research
Abstracts - 2006
|
|
Removing Camera Shake from a Single PhotographRob Fergus, Barun Singh, Aaron Hertzmann, Sam Roweis & William FreemanIntroductionCamera shake during exposure leads to objectionable image blur and ruins many photographs. Conventional blind deconvolution methods typically assume frequency domain constraints on images, or overly simplified parametric forms for the motion path during camera shake [1]. Real camera motions can follow convoluted paths, and a spatial domain prior can better maintain visually salient image characteristics. We introduce a method to remove the effects of camera shake from seriously blurred images. The method assumes a uniform camera blur over the image and negligible in-plane camera rotation. In order to estimate the blur from the camera shake, the user must specify an image region without saturation effects. Figure 1. shows a typical image, corrupted by camera shake; the result of Photoshop's "Unsharp Mask" operation on the image and the result of our algorithm on the image.
MethodWe have developed promising methods for removing the effects of camera shake, having two key improvements over previous work. First, we exploit recent research in natural image statistics, which shows that photographs of natural scenes typically obey very specific distributions of image gradients [2]. Second, we use a Bayesian approach [3] that takes into account uncertainties in both the blur kernel and the estimated image, allowing us to find the image implied by a distribution of probable blur kernels, reducing reconstruction artifacts. The method first computes a distribution of possible motion blurs caused by the camera shake, before estimating a plausible and pleasing image reconstruction consistent with those probable blurs. For efficiency, we currently run the inference procedure on a user-selected sub-window of the image. This gives us a blur kernel which we then use with a standard non-blind deconvolution method, Richardson-Lucy [4], to unblur the whole image.Future workWe are currently working to improve the quality of the Richardson-Lucy reconstruction, by introducing image priors into the scheme. Additionally, we are looking to incorporate various non-linearities which are currently not modeled, such as saturation, into the inference procedure.ResultsFigures 2 and 3 show some more examples of our algorithm on real images submitted by various people.
References:[1] D. Kundur and D. Hatzinakos. Blind Image Deconvolution. In IEEE Signal Processing Magazine, Vol. 13 , No. 6, May, 1996. [2] B. Olshausen and D. Field Emergence of simple-cell receptive field properties by learning a sparse code for natural images. In Nature, pages 607-609, vol. 381, June, 1996. [3] J. Miskin and D. MacKay. Ensemble Learning for Blind Image Separation and Deconvolution. In Advances in Independent Component Analysis, Springer-Verlag, 2000. [4] W. Richardson. Bayesian-based iterative method of image restoration. In Journal of the Optical Sociatey of America A , pages 55-59, vol. 62, 1972. |
 |
|||||||||
|
