基于字典学习的残差信息融合图像去噪方法

摘要

传统去噪算法只考虑从含噪图像中恢复出图像信息，然而对去噪后残差信号的利用却并未加以重视。针对图像去噪后残差信号中包含有用信息的特点，提出了一种基于字典学习的残差信息融合图像去噪方法。首先使用字典学习方法对单幅含噪图像进行去噪；然后对首次降噪后的残差图像进行图像块筛选；再对筛选出的图像块再次进行去噪处理；最后在小波域实现两幅图像的融合得到最终的去噪图像。实验结果表明，与传统基于字典学习的去噪方法相比，所提方法能够进一步提取残差信号中的图像特征信息，在峰值信噪比和结构相似度上都有所提升。特别是对一些细节较为复杂的场景图像，具有更好的去噪效果，从而证明了残差信号对于图像去噪的重要作用。%The traditional image denoising algorithms focus on how to restore image information,but it pays less attention to the effects of residual signals obtained after denoising.Since residual signals contain the useful information, a dictionary -learning -based image denoising algorithm using information fusion of residuals is proposed in this paper.This algorithm first applies the traditional dictionary -learning -based denoising approach on a noisy image.Then,some image patches are selected from the residual image obtained in the previous step.These image patches are further denoised. Finally,the denoised image obtained in the first step and the one attained by the residual image are fused in the wavelet domain.The experimental results show that,compared with traditional dictionary -learn-ing -based denoising algorithms,the proposed approach can consequently extract feature information of images from the residual signals and improve denoising performance with respect to peak signal -to -noise ratio (PSNR)and structural similarity index measurement (SSIM).The proposed algorithm is especially suitable to images containing complex scenario details and can achieve better denoising performance,which illustrates the important effects of residual signals for image denoising.