Designing the optimal structure of a neural filter

In this paper, we propose a new method for designing the optimal structure of a neural filter (NF), and evaluate its performance. In order to verify the validity of the proposed method, we apply the proposed method to the NF that has been trained to achieve the function of a linear filter with kernel of known shape, and verify that the optimal kernel is achieved. The experimental results demonstrate that the optimized NF by the proposed method achieves the optimal generalization ability, and the performance of the optimized NF by the proposed method is superior to that of the original NF. By the comparative evaluation with the NF that has been trained to reduce noise in medical images, we show that the proposed method is superior to the ability of the conventional method quantitatively in terms of the performance of optimization, the filter performance of the optimized NF, and the generalization ability of the optimized NF