Image denoising algorithm of U-shaped network based on residual neural network and wavelet transform
This is only used to test our algorithm.
Recent deep learning algorithms have shown excellent performance in fast imaging and image restoration tasks, in pursuit of a stable deep learning neural network model and improving image denoising performance in Gaussian noise environments. In this work, a residual residual block with residual algorithm and a multi-level wavelet transform neural network model with symmetric structure are proposed to solve the image denoising problem. The algorithm network effectively simplifies the classical U-shaped structure symmetry. The hierarchical and complex instability of the neural network model reduces the computational redundancy cost of the model, and customizes the residual module unit according to the characteristics of the residual network to prevent the gradient disappearance and difficult optimization problem of the network model. The stability of the neural network model is used. The proposed network algorithm uses a multi-level wavelet transform algorithm to downsample the convolution in other classical neural networks, which avoids the problem of using the downsampling convolution to lose the image feature details. In addition, this paper uses two training samples with different sizes to verify the experiment. The data results show that the performance of the proposed algorithm model under the larger training data samples has been improved. Therefore, the proposed algorithm inherits the advantages of the classical neural network structure and the classical algorithm while making up for the shortcomings of the algorithm. The experimental data proves that the proposed algorithm model improves the model in a high noise environment to some extent. Under the image denoising performance, the higher quantitative peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) evaluation indicators are obtained, and the detailed feature information of the image is better preserved to achieve better visual effects.
Key words: deep learning; image restoration; image denoising; wavelet algorithm; residual network