An adaptive filtering interpolator using neural networks

Filtering interpolators presented by Lucke and Stocker (1993) have advantages in reducing interpolation error in image background clutter-suppression systems especially for data with low sampling rates. Before they are to be applied, a fixed parameter α should be predetermined. The authors think if the parameter α is well adjusted, it may also be useful to recover an image from a less densely sampled image. Experiments show that interpolation error relies greatly on the parameter α and the best values of α for certain images are much different. Therefore, how to determine the values of α becomes the key problem for this application. In this paper, the authors develop a neural network based adaptive system to automatically adjust the value of α. A modified robust BP algorithm is used in the training procedure for the authors´ special use. Simulation results show that α can be generated automatically by the neural networks instead of being blindly predetermined to a fixed value. Compared to the interpolator with best fixed parameter α, interpolation results are also improved