MLP-based Image Interpolation Using Local Characteristic of Wavelet Coefficients

Image interpolation in the wavelet domain is the estimation problem of the finest detail coefficients. A wavelet coefficient has an interscale dependency and its Liptschitz exponent is found to be different according to the energy of the coefficient. This implies the possible existence of functional mapping from one scale to another. If we can get the mapping parameters from observed coefficients, it is possible to predict the finest detail coefficients. In this paper, we exploit the multi-layer perceptron (MLP) to learn the mapping from the coarser scale to the finer scale. Phase uncertainty makes it difficult for the MLP to learn the interscale mapping. We solve this location ambiguity by using a phase-shifting filter. In the simulation results, we show that the proposed scheme outperforms the previous wavelet-domain interpolation method as well as the conventional spatial domain methods.