| 贝叶斯-EM联合建模的图像修复与噪声抑制 |
投稿时间：2026-06-04 修订日期：2026-07-15  点此下载全文 |
| 引用本文： |
| 摘要点击次数: 19 |
| 全文下载次数: 0 |
|
| 基金项目:本文系2024年广东省普通高校特色创新类项目“基于智能检测的无线降频技术物联网研究”(项目编号：2024KTSCX368)成果之一 |
|
| 中文摘要:摘 要针对图像修复与噪声抑制分离处理导致的效率低、误差累积问题，提出一种基于贝叶斯联合建模与期望最大化(EM)优化的图像修复与噪声抑制统一方法。通过构建融合图像先验、噪声模型及缺失机制的概率模型，将双任务转化为统一的最大似然估计问题，并采用EM算法协同求解。在随机缺失与块状缺失叠加噪声的多样场景下开展验证实验，结果表明，相较Transformer和CNN等传统方法，所提方法在图像修复与去噪性能上均能取得明显提升，峰值信噪比(PSNR)与结构相似度(SSIM)平均提升超11.3%与8.5%，且对大缺失区域与强噪声情况具有更强的鲁棒性。 |
| 中文关键词:图像修复 噪声抑制 样本缺失 混合高斯模型； 贝叶斯理论 |
| |
| IMAGE INPAINTING AND NOISE SUPPRESSION VIA BAYESIAN-EM JOINT MODELING |
|
|
| Abstract:AbstractTo overcome the inefficiency and error accumulation caused by separate image inpainting and noise suppression, this paper proposes a unified method based on Bayesian joint modeling and Expectation Maximization (EM) optimization. A probabilistic model integrating image prior, noise, and missing mechanisms is built, transforming the dual-task into a unified maximum likelihood estimation problem solved collaboratively via the EM algorithm. Experiments under diverse scenarios (random/block missing with noise) show that compared to state-of-the-art methods (e.g., Transformer, CNN), our approach achieves superior inpainting and denoising performance, with PSNR and SSIM improved by over 11.3% and 8.5% on average, while exhibiting stronger robustness to large missing regions and high noise levels. |
| keywords:Image inpainting Noise suppression Missing samples Gaussian mixture model Bayesian theory |
| 查看全文 查看/发表评论 下载pdf阅读器 |
位访问者,今日访问
