Abstract: Many of the Patch based image denoising algorithms filter overlapping image patches and aggregate multiple estimates for the same pixel via weighting. Current weighting and filtering approaches assume the restored estimates as independent random variables, which is not convenient with the reality. In this paper, we consider the correlation among the estimates and propose model to estimate the Mean Squared Error under various weights of the image patches. This model identifies the overlapping information of the patches, and then use the optimization to try to minimize the MSE. We propose a new weighting approach based on Quadratic Programming, which can be embedded into various denoising algorithms.

Keywords: Patch based, Mean Squared Error (MSE), Weighting, Bias variance model, Filtering, Weighting.