Applications of Neural Networks in Video Signal Processing

Although color TV is an established technology, there are a number of longstanding problems for which neural networks may be suited. Impulse noise is such a problem, and a modular neural network approach is pre(cid:173) sented in this paper. The training and analysis was done on conventional computers, while real-time simulations were performed on a massively par(cid:173) allel computer called the Princeton Engine. The network approach was compared to a conventional alternative, a median filter. Real-time simula(cid:173) tions and quantitative analysis demonstrated the technical superiority of the neural system.