scatter-plot method for blind additive noise variance evaluation in images

V.V. Abramova

A method for blind additive noise variance evaluation has been proposed. It is based on first degree polynomial fitting into “normal” reference points on scatter-plot of local variance and mean estimates. Numerical simulation results obtained for images from TID2008 database show that accuracy of the method is higher in comparison to other known methods. The most significant accuracy increase is observed for highly-textural images containing many edges and small objects. The method is effective both for cases of spatially uncorrelated and correlated noise.

Key words: remote sensing, additive noise, blind variance evaluation, regression analysis.