Research On Image Restoration Algorithm Based On Deep Convolutional Neural Network

Image super-resolution and image denoising have attracted much attention in digital image restoration and are widely used in computer vision,such as community security and satellite maps.They can be considered as the problems of image repairing that makes the images clear.There are some differences between image super-resolution and image denoising.Image super-resolution is the process of raising a low-resolution image to a high-resolution image.Image denoising is to remove noise that does not belong to a picture.Different image repairing algorithms are used for image super-resolution and image denoising.Although there are a number of image restoration algorithms,performance also should be further improved,especially for restoration a single image.In recent years,deep convolutional neural networks have made breakthroughs in modern digital image processing.Compared to traditional algorithms,deep convolutional neural networks achieve superior performance on a series of challenging image processing problems such as image classification and target detection.This is because of their excellent self-learning abilities.Deep convolutional neural networks learn through a large number of training samples,obtain relevant information within the image,and then use the information to achieve specific functions.Deep convolutional neural networks also have excellent performance in image restoration.This thesis focuses on image super-resolution and image denoising.The main contributions of the thesis include:(1)A novel cascaded deep convolutional neural network model is proposed for image super-resolution.In order to improve the performance of the network model,a multi-scale feature mapping network structure is proposed to replace the original convolutional layer.The structure of this multi-scale feature map enables the network to use multiple convolution kernels of different sizes in the same layer of convolution.In addition,the model also incorporates residual learning,network cascading and parallel strategies.The proposed single network model has the ability to repair image super-resolution with different magnification factors.A detailed performance analysis of the network model is then performed and validated by a large number of experiments.(2)A deep convolutional neural network model is designed for image denoising.Because noise often contains less information,the designed network model has the ability to directly separate the noise mask from the noise-contaminated image.Then,the desired clear image can be achieved by the difference of the noise-contaminated image and the noise mask.The designed network model also has the ability of a single model to remove different levels of noise.That is to say,only one network model can be used to simultaneously repair different contaminated images with different noise levels.During the training process,we add the noise to the training data in a way that does not have a fixed noise distribution.Experimental results show that the designed network model has better denoising performance than other algorithms.