Gaussian Noise Filtering Using Pulse-Coupled Neural Networks

In the process of collecting or transmitting images, various noise interferences are often introduced, especially in a multi-image sensor network, and noise has an important influence on subsequent image processing. Gaussian noise is a common noise in such systems. In order to filter Gaussian noise better, the neighborhood gray level difference weight matrix is proposed and applied to the Pulse-coupled neural network (PCNN). The matrix corresponds to the coupling-connection matrix of the PCNN and is determined by the related constraint relationship. From the perspective of image pixels, the neighborhood gray level difference weight matrix can adaptively change the gray level of the noisy pixels in the center of the neighborhood and improve the correlation of pixel gray levels in the neighborhood. From the macro perspective, the introduction of the neighborhood gray level difference weight matrix converts the image denoising process into a two-dimensional convolution operation. When the initial conditions are determined, parallel processing can be realized, which greatly improves the efficiency of the algorithm. These advantages make the proposed algorithm can be better combined with CNN and other networks as the front-end denoising module of these networks. The specific experiments show that the denoising effect of this algorithm is better, especially under the higher variance Gaussian noise.