Recursive filtering a video image using activity mapping

摘要

A method of recursive filtering a video image includes determining a local variance sigma² in luminosity in the vicinity of a picture element on the image; making an estimate of a value of noise variance sigma_r²; obtaining a surface for value of a proportional parameter K from the equation $K=\mathrm{alpha}\left[\frac{\mathrm{rho}.\mathrm{lambda}.{\mathrm{sigma}}^2}{\mathrm{rho}.\mathrm{lambda}.{\mathrm{sigma}}^2+\mathrm{tau}.{\mathrm{sigma}}_r^2}\right]+\mathrm{beta}$ where 1<=sigma<=128, 1 <=sigma_{r <=64 and rho, tau, alpha, and beta are empirical constants, and selecting a value of lambda to scale a relative contribution to the value of K by the luminosity variance and the noise variance. The image is recursively filtered using the proportional parameter K to sum proportions of the current image and an immediately preceding image.}