摘要
针对现有卷积神经网络在处理高密度椒盐噪声图像时易出现边缘模糊与细节丢失的问题,文章提出了一种融合模糊逻辑与深度学习的图像去噪算法。该算法引入了一个可学习的软模糊掩码模块,能根据输入图像逐像素估计噪声置信度,并结合改进的深层U-Net网络以增强图像特征提取与去噪能力。此外,为提升复原图像的结构保真度与边缘清晰度,设计了融合均方误差、结构相似性(SSIM)与边缘损失的混合损失函数。实验结果表明,该算法在主观视觉效果与客观评价指标上均展现出优越性能。
To address the issues of edge blurring and detail loss in existing convolutional neural networks when processing high-density salt-and-pepper noise images,this paper proposes an image denoising algorithm that integrates Fuzzy Logic Deep Learning.The algorithm introduces a learnable soft fuzzy mask module to estimate noise confidence pixel by pixel based on the input image,and combines an improved deep U-Net network to enhance image feature extraction and denoising capabilities.Additionally,to improve the structural fidelity and edge clarity of the restored image,this paper designs a hybrid loss function incorporating Mean Squared Error,Structural Similarity Index Measure(SSIM),and edge loss.Experimental results show that the algorithm exhibits superior performance in both subjective visual effects and objective evaluation metrics.
作者
田汇原
张小波
TIAN Huiyuan;ZHANG Xiaobo(Xi'an Shiyou University,Xi'an 710065,China;Xianyang Normal University,Xianyang 712000,China)
出处
《现代信息科技》
2026年第2期134-138,144,共6页
Modern Information Technology
