基于改进U-Net神经网络的图像去噪算法

Image Denoising Algorithm Based on Improved U-Net Neural Network

针对目前常见的U-Net网络结构以及现有的图像去噪算法在去除图像噪声时,处理后得到的图像较为模糊且图像的边缘纹理过于光滑缺乏真实性的问题,提出了一种改进的U-Net网络结构去噪算法。它由去噪模块以及边缘信息提取模块组成,首先,利用U-Net++中的跳跃连接应用到原始的U型去噪子网中,密集连接的U型去噪网络可以减少编码器与解码器特征映射之间的语义差距,还原出更清晰的图像。其次,基于VGG-16网络结构的边缘信息提取模块对去噪网络处理后的图像进行特征提取,同时反向优化U型去噪模块,还原出更真实的图像。实验表明,在常见的Set5、Set12、Kodak24和CBSD68数据集测试所提出的算法,在图像的客观评价指标上均优于目前具有代表性的去噪算法,同时图像的边缘细节和纹理特征更清晰真实,视觉效果上更好。

Aiming at the problems that the current common U-Net network structure and existing image denoising algorithms remove image noise,the processed image is blurry and the edge texture of the image is too smooth and lacks authenticity.An improved U-Net network structure denoising algorithm was proposed.It consists of denoising module and multi-feature fusion edge information extraction module.Firstly,the jump connection in U-Net++was applied to the original U-shaped denoising subnet.Densely connected U-shaped denoising network can reduce the encoder semantic gap between the feature map and the decoder feature map restores a clearer image.Secondly,the edge information extraction module based on the VGG-16 network structure performed feature extraction on the image processed by the denoising network.The U-shaped denoising module was reversely optimized to restore a more realistic image at the same time.Experiments show that testing the algorithm proposed on the common Set5,Set14,Kodak and McMaster data sets is better than the current representative denoising algorithms in the objective evaluation of images.The edge details and texture features of the image are clearer and more realistic,and the visual effect is better.