Adaptive threshold selection of anisotropic diffusion filters using spiking neural network model

Image denoising takes place as the first step in many image processing operations. Anisotropic diffusion filters (ADFs) have remained popular for years because of their ability to preserve edges, along with their success in denoising. One of the biggest disadvantages of ADFs is that the threshold value, which has a significant effect on the success of denoising, is chosen by the user. Many techniques based on image gradients have been developed to determine the threshold value optimally. However, it is a big handicap that the gradient value is very sensitive to noise. Recent trends in bio-inspired research, particularly in the context of the human visual system (HVS) and Spiking Neural Networks (SNNs), have demonstrated their efficacy in edge detection. In this paper, a novel approach employing a conductance-based Integrate and Fire neuron model is proposed for fully adaptive threshold selection. The proposed method's effectiveness is evaluated using a dataset of 600 noisy images generated from 60 COVID-19 computed tomography (CT) scans, incorporating Additive White Gaussian Noise (AWGN). Experimental results, including the structured similarity index (SSIM) and peak signal-to-noise ratio (PSNR) values, indicate that the threshold values determined by the proposed method yield superior denoising outcomes compared to existing fully adaptive threshold selection techniques.