| Digital image processing has always been a research hotspot in the field of computer vision.Among them,image denoising is one of the most basic and common problems in digital image processing.As people’s requirements for image quality continue to increase,how to remove noise from images has become a very important task in many image processing applications.Traditional image denoising methods are mainly based on smoothing filtering algorithms,which attenuate the noise by smoothing the image,but this denoising way is prone to losing the detailed information of the image,resulting in a decrease in image quality.In contrast,image denoising algorithms based on deep learning can better improve image denoising performance and reduce detail loss,thereby emerging as a new trend in image denoising technology.With the continuous development and improvement of deep learning technology,more and more deep learning-based image denoising algorithms have been proposed,among which convolutional neural networks(CNNs)are the most widely used.It is worth noting that,some CNN-based image denoising algorithms still have problems such as overfitting and insufficient generalization,which limit the application scope and performance of these algorithms.To better solve these problems,a pixel attention-guided CNN-based image denoising algorithm PAGDNet is proposed in this dissertation,which mainly integrates asymmetric convolution,pixel attention mechanism,and residual learning techniques to improve image denoising efficiency and performance.Specifically,asymmetric convolution is adopted to reduce the network parameters for reducing the computational costs,thereby improving the image denoising processing efficiency and reducing the occurrence of overfitting of the proposed algorithm,to a certain extent.The proposed algorithm is guided to pay more attention to the multi-noised pixels and high-frequency image areas of the image by the pixel attention mechanism,allowing it to better extract more noise information from the noisy image,so as to achieve the purpose of improving the image denoising performance.Residual learning is combined to better address the training difficulties of deep neural networks,while preventing gradient explosion and vanishing,thus improving the training efficiency and performance of the proposed algorithm.Simulation experiment results indicate that,compared with most of the advanced denoising algorithms,the proposed image denoising algorithm PAGDNet achieves better denoising performance in three major denoising tasks of processing images with synthetic noise,images with blind noise,and color images with real noise.Additionally,the proposed image denoising algorithm PAGDNet is strongly competitive in terms of image denoising processing efficiency. |
