Abstract: Embodiments of the present invention provide for thresholding data elements (such as coefficients). According to one embodiment, a number of functions (including feature detector functions) can be applied to a data element to generate responses. Additionally, noise profiles can be consulted to determine an expected result for each function if the data element represents noise. An output value (e.g., after thresholding) for the data element can be generated based on the actual responses of the functions and the expected responses.
Abstract: Embodiments of the present invention provide a method and system for characterizing noise in a set of data based on an analysis of subsets of the data. For example, noise for an image can be characterized based on analyzing selected regions of the image. The characterization of the noise can be stored in a data structure that models the noise as jointly dependent on values of multiple parameters (pixel values for multiple channels). The noise can be further characterized for one or more wavelet subbands of the set of data.