Research on Image Denoising Algorithm Based on Self-Supervised Learning

In recent years, neural network models based on self-supervised learning have shown excellent performance in the field of image denoising. However, current denoising methods either rely on specific noise models or can only perform denoising tasks under specified noise distributions. Moreover, these network models often have a large number of parameters, leading to inefficient parameter usage. To address these issues, inspired by the traditional Partial Differential Equation (PDE) methods in image denoising, we attempted to combine Neural Ordinary Differential Equations (NODEs) with Convolutional Neural Networks (CNNs) to produce a neural network model with NODE-like properties. This was aimed at improving the parameter efficiency of the model. Additionally, we introduced a new architecture within the model, namely normalization equivariance, which enables existing neural network models to possess the property of normalization equivariance. Compared to models lacking this property, models with normalization equivariance often exhibit stronger generalization capabilities and robustness. Through comparative experiments conducted on artificially synthesized noise and noise from real-world colored images, the results indicate that our proposed method demonstrates remarkable denoising effects across different scenarios. It exhibits stronger generalization capabilities and restores most of the texture details within the original images.