Adaptive Threshold Selection in Image Smoothing Problems Using a Sigma Filter

Abstract

The method for obtaining an estimate of random component (noise) variance in an image is proposed. Noise variance estimation is used in smoothing sigma filter to form a cutoff threshold for pixels with brightnesses exceeding the brightness of the central pixel in the window. Correct setting of the cutoff threshold allows to save the boundaries of the brightness difference and also to provide noise suppression. Noise variance estimation is performed using a horizontal polynomial filter, which provides an unbiased estimation of polynomials up to third-degree inclusively and, as a result, obtains the correct variance estimation.