Structure-adaptive evaluation of additive noise level in images
Iryna B. Ivasenko, Roman M. Palenichka · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1998
In the proposed paper, the problem of noise evaluation is considered with application to image filtering and segmentation. The underlying structural model of original image is considered which describes the shape of image objects or their parts. The distinctive feature of the presented model is the separate modeling of object's planar shape as well as the image intensity function. For the intensity function model of original image, a piecewise polynomial model of low degrees (up to the second one) is considered. Then, noise to be evaluated is treated in a broad sense, namely as the intensity residuals of the piecewise polynomial modeling. It is also assumed that most of the pixels satisfy the polynomial model except for a relatively small number of edge points between homogeneous regions and fine details. A robust noise variance estimator is proposed for the images corrupted by outliers, i.e. impulsive noise.