Image denoising method based on expansion convolution

Abstract

The invention relates to an image denoising method based on expansion convolution. The method specifically comprises the following steps of 1, preparing training data; 2, building a model; 3, compiling an image acquired in the step 2 to acquire a compiled model; 4, adding additive white Gaussian noise on the block image in the step 1 to acquire batch pictures with noise; 5, training the batch pictures with noise acquired in the step 4, thus acquiring a trained model; 6, preparing test data; and 7, importing the acquired test image into a prediction function of the model acquired in the step 5to acquire the denoised image. According to the denoising method provided by the invention, the sharp edge and fine details can be recovered, and the enjoyable visual effect can also be generated in the smooth area; furthermore, the network structure in the method is composed of 14 layers, the needed time can be reduced, and the efficiency can be improved.

Description

technical field [0001] The invention belongs to the technical field of image denoising, and in particular relates to an image denoising method based on dilated convolution. Background technique [0002] Image denoising refers to the process of reducing noise in digital images. In reality, digital images are often affected by imaging equipment and external environmental noise interference during digitization and transmission, which are called noisy images or noisy images. [0003] DnCNN uses residual learning to remove the clean image in the hidden layer to obtain a noisy image, and then subtracts the noisy image from the noisy input image to obtain a clear image. DnCNN uses a 17-layer network, of which the first layer is expansion convolution + nonlinear activation function (Relu), the 2nd-16th layer is expansion convolution + batch normalization + nonlinear activation function (Relu), and the 17th layer is Expansion convolution, the number of layers of this type of network...

AI Technical Summary

Problems solved by technology

DnCNN uses a 17-layer network, of which the first layer is expansion convolution + nonlinear activation function (Relu), the 2nd-16th layer is expansion convolution + batch normalization + nonlinear activation function (Relu), and the 17th layer is Expansion convolution, the number of layers of this type of network is too deep, it takes a long time, so it will reduce the efficiency

Images