Adaptive Deblurring of Multi-Layer Microscopic Images with Single-Layer Cellular Neural Networks

In this paper we investigate a possible way of deblurring the digital images of thick microscopic slices with parallel networks. Images obtained by microscopy typically suffer from two effects: the final resolution of the optical system and there are errors due to the final focal depth. Errors can be modelled with the convolution of the ideal images. To restore these values deconvolution can be applied in space which is a very fast function of Cellular Neural Networks (CNNs). That is why we use CNNs to solve the problem, while during the solution the neighbouring microscopic slices are also taken into consideration in the restoration. Parameters needed for deblurring are obtained by either well-known optical computations or genetic algorithms. In this paper we show that the deconvolution of images representing the layers of thick microscopic slices can be approximated with a one layer structure with minimal loss of precision. This means that much simpler structures can be applied for the computations in VLSI hardware realisations or computation times are reduced significantly when using software simulations.