Image Denoising Algorithm Based On PDE And Deep Learning

Among various tasks of digital image processing,image denoising is one of the basic and important tasks.In recent decades,scholars have conducted in-depth research on digital image processing from various angles,and proposed a large number of image restoration algorithms.However,due to the diversity and complexity of image information,whether it is a traditional image denoising method or a deep learning denoising algorithm,there are still many difficulties and work to be done.This paper mainly studies the image denoising algorithm based on PDE and deep learning.The main contents are as follows:1.Method for image denoising based on total generalized variation and wavelet threshold modelWavelet image denoising model could blur the image edges,so we propose an image denoising algorithm based on a combination of the total generalized variation model and the adaptive wavelet threshold model.First of all,the wavelet transform is used to decompose the noisy image to obtain the wavelet coefficients of each dimension.Then the low-frequency coefficients are processed by the second-order total generalized variational model,and the high-frequency coefficients are processed by the improved adaptive threshold model.Finally,we reconstruct the image.Through numerical experiments on images with different noise levels,visual observation and objective evaluation indicators show that the denoising algorithm in this paper has indeed improved.2.Method for a new auto-encoders convolutional neural network image denoising algorithmIn order to improve the efficiency of image denoising,we propose a new autoencoders convolutional neural network image denoising algorithm.This new algorithm changes the activation function of the convolutional layer and the skip connections of the whole network,and it has improved the selection of the learning rate.The new model has simpler structure design and fewer number of convolutional network layers.Compared with the original model,numerical experiments show that the new model has the characteristics of less running time,better visual effects,and larger peak signal-tonoise ratio.