The technology herein describes a methodology and algorithm for removing blur effects from photographs affected by camera shake. This is achieved by creating a “blur kernel” for the affected picture, which is a model of the motion of the camera that caused the resultant blur. Due to the complexity and irregularity of the associated camera motion, the construction of the blur kernel requires a number of assumptions and estimations in order to be used as a suitable guide for deblurring. The methodology that guides the judgement of the necessary assumptions and estimations is what sets this technology apart from existing techniques.
In this case, the system makes use of two central methodological improvements to guide the construction of the blur kernel. First, the system takes advantage of recent research in image statistics that suggests pictures taken of natural scenes typically follow very specific distributions of image gradients. This serves as a very useful tool for estimation. Second, by analysis of distributions of probable images, the technique takes into account the uncertainties associated with the unknown camera movement. Once the blur kernel is constructed, it can be used with a standard deblurring algorithm to produce a corrected version of the desired picture.