An image reconstruction system by neural network with median filter
Y. Chigusa, Keita Suzuki, T. Hattori, M. Ikegami, Mamoru Tanaka · 1993 IEEE International Symposium on Circuits and Systems · 2002
The authors describe a novel image reconstruction system with a median filter from halftoning images based on a neural network. The massively sparse Hopfield neural network is applied to this system. The thresholding function is modified to the median filter, which outputs the selected signal. When a condition is satisfied, this system converges. For the implementation, only binary-weighted synapses are employed, where the synaptic weight falls off at the inverse of the distance between neurons. Therefore the conductance matrix is relatively sparse. Two main advantages of this dynamic system are clearly established. The method achieves both image smoothing and edge-holding. Simulation results show that the method produces a more natural image for reconstruction.>