Design of complex-valued CNN filters for medical image enhancement

In this paper, we present a new image enhancement technique, using cellular neural network (CNN) filters with complex weighting factors, that is applicable to medical images. Since CNN-type filters have only spatially local interconnections and the number of connections between neurons is relatively low, the required computation in the learning phase is a reasonable amount. However, the output/input behavior is restrictive. The proposed CNN filters are designed as complex-coefficient filters which can improve the output SNR and process the 2D analytic signals of input images. The filter parameters are determined by applying a complex domain backpropagation algorithm. Through several simulations, it is shown that the proposed filters are robust and noise-tolerant for medical images