Image restoration using parametric partial projection filter

The partial projection filter has the following features: (a) when there is no noise, the method restores the best approximation of the images belonging to a certain subspace of the original image space; (b) when there is a noise, the method restores the image which minimizes the variance in regard to the best approximation of (a). In addition, by slightly relaxing the condition (a), the variance in (b) can drastically be reduced. This paper proposes a method where the variance in (b) can be reduced without reducing greatly the capability of the partial projection filter in (a). The proposed restoration filter is called a parametric partial projection filter, and its usefulness is verified for a motion-blurred image.