A Spline-Based Framework for Perfect Reconstruction AM-FM Models

For the first time, we present a multicomponent perfect reconstruction AM-FM image model that is fully consistent with human visual perception. We design a separable perfect reconstruction wavelet filterbank based on the discrete Coiflet and implement it in a maximally decimated parallel structure characterized by excellent joint space-frequency localization. Using the zeros in the channel response magnitudes, we further decompose each channel into a sum of highly localized, non-separable orientation selective sub-channels. We fit the responses of these sub-channels with nonlinear spline models and apply error-free continuous demodulation algorithms to obtain a new perfect reconstruction AM-FM model. This is significant because the new model could be used as the basis for a new theory of modulation domain image processing