We address the question of noise reduction on images of very low SNR that predominantly contain quasiperiodic image features. In this case neither the Fourier transform nor wavelet transforms are appropriate signal representations that allow for discrimination of signal and noise in the tranform domain. We show the construction of a multidimensional,cycle-spinning based, translation- and rotation-invariant waelet packet transform and employ this transform is a scheme for denoising data with an unknown spectral signal-noise relationship. We compare the signal reconstruction performance with that of other noise reduction methods. The proposed scheme allows for data exploration on low-SNR imagery of the given type.
展开▼