Regularized image reconstruction using neural networks
Ronald J. Steriti, Michael A. Fiddy · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1992
Iterative methods have long been studied in order to reconstruct images from limited noisy spectral data or low pass filtered noisy images; they rely on minimizing a well-defined energy function. Such methods can be implemented on Hopfield neural networks, as a direct result of comparing energy function parameters. Consequently, a fully parallel (neural) processor can be programmed to implement a reconstruction algorithm. We have studied the properties of these neural solutions and show that they provide a regularized and apodized result with some attractive and interesting properties.