Time-frequency distribution inversion of the Radon transform

In using filtered backprojection to compute the inverse Radon transform, the Hilbert transform-derivative filtering operation amplifies noise. Filtering the noise generally reduces resolution and smoothes edges. It would be desirable to spatially filter the noise where the projections do not have high-frequency components, and accept the noise where high-frequency components are present, since these must be preserved for the Hilbert transform-derivative. The short-time Fourier transform is used in applying a time-frequency mask filter that zeros out each projection in locations where local signal energy is below a threshold. Results show improvement over reconstruction using time-invariant smoothing filters.<>