Image noise filtering has been widely perceived as an estimation problem in the spatial domain. We deal with it as an estimation problem in an uncorrelated transform domain. This idea leads to a generalization of the adaptive linear minimum mean square error (LMMSE) estimator for filtering noisy images. In our proposed method, the transform-domain local statistics obtained from the noisy image are exploited. Due to the fact that the transform-domain local statistics carry more information about the image than the spatial-domain local statistics do, improvement in noise filtering is gained overall and is particularly significant in the vicinity of edges.