A parametric technique for superresolution image reconstruction

A model-based approach for superresolution signal reconstruction from noisy bandlimited Fourier data is proposed. The approach combines the virtues of parametric and nonparametric techniques by maximum-likelihood fitting of the data by a mixture of exponentials and smooth basis functions. Performance bounds are derived and analyzed, and a model computationally efficient algorithm is described. The performance of the algorithm in simulations closely matches the bounds over a wide range of operating conditions. Overall, the reconstructions are far superior to those obtained by traditional Fourier transform processing.<>