Minimum cross-entropy noise reduction in images
Robert F. MacKinnon · Journal of the Optical Society of America A · 1989
A method of noise reduction is described that reduces random noise in images through cross-entropy representation under simple constraint bounds placed on linear orthogonal transform variables. The bounds depend on the noise statistics, which must be estimated independently, and on prior knowledge. The bounds may be adjusted through use pf a so-called tightness parameter. In practice, solutions represent a compromise between the noisy image and the prior knowledge for which the tightness parameter governs the reduction in the noise variance. The role of the knowledge is illustrated by using two examples, one simple and one complicated. Results based on Fourier prior are presented. Examples of speckle noise reduction for synthetic aperture radar images of and Walsh transforms the ocean surface are given as illustrations of a practical application.