基于DnCNN-IID的RAW图像盲降噪算法
Blind denoising algorithm for RAW images based on DnCNN-IID
李博 1赵一诚 1丁辉2
作者信息
- 1. 首都师范大学信息工程学院,北京 100048
- 2. 首都师范大学信息工程学院,北京 100048;成像技术北京市高精尖创新中心,北京 100048
- 折叠
摘要
由于RAW图像中存在整体偏暗和对比度不足,以及亮度不一致等问题,导致RAW图像中的噪声复杂多样.提出了一种改进的降噪模型DnCNN-IID(denoising convolutional neural network with image inversion and down-sample,DnCNN-IID)抑制RAW图像中的复杂噪声,增强图像质量.该模型基于DnCNN网络,通过对图像进行反通道处理进行图像增强,增强图像对比度,突出图像中的细节和边缘,同时突出噪声的特征信息;通过加入图像下采样使网络处理效率提升的同时,扩大了网络感受野,提高模型对输入图像的全局信息的感知能力,更加有效的抑制噪声.为了验证算法的有效性,在BSD500 数据集、Ex/600 数据集与RAW数据集上与主流方法进行了比较,实验结果表明,所提模型在PSNR、SSIM、MSE多个评价指标上得到较好的提升.
Abstract
Due to the overall dark,insufficient contrast,and inconsistent brightness in RAW images,the noise in RAW images is complex and diverse.In this paper,an improved Denoising model DnCNN-IID(Denoising Convolutional Neural Network with Image Inversion and Down-sample)is proposed to suppress complex noise in RAW images and enhance image quality.The model is based on DnCNN network,and the image was processed by the image inversion for data enhancement to improve the image contrast,highlight the details and edges in the image,and highlight the characteristic information of the noise.By adding image down-sampling,the network processing efficiency is improved,the network receptive field is expanded,the model's ability to perceive the global information of the input image is improved,and the noise is suppressed more effectively.To verify the effectiveness of the algorithm,the proposed model is compared with the mainstream methods on the BSD 500 dataset,Ex/600 dataset and RAW dataset.The experimental results show that the proposed model has a better improvement in PSNR,SSIM and MSE.
关键词
RAW图像/卷积神经网络/盲降噪/暗图像增强Key words
RAW Image/convolutional neural network/blind noise reduction/dark image enhancement引用本文复制引用
出版年
2024
NETL
NSTL
万方数据