Abstract: Different from single-domain image denoising, dual-domain image denoising effectively combines two sorts of algorithms working in their respective domains together for preserving image details and suppressing noise simultaneously. However, the variance of image is a prerequisite for the dual-domain image donoising, which is hard to be determined accurately in a practical application. As for the dual-domain filtering constructed specifically on the basis of bilateral filtering and short-time Fourier transform (STFT), this paper carries out three aspects of work: (1) Estimating the noise image by subtracting dual-domain denoised image to noisy image; (2) Estimating the variance of non-stationary noise on the basis of weighting coefficients of bilateral filtering; (3) Adapting dual-domain filtering to the case of non-stationary noise. The experimental results and related quantitative analysis demonstrate that the proposed adaptive dual-domain filtering can more effectively suppress image noise while retaining image details.

Key words: image denoising, bilateral filtering, short-time Fourier transform, dual-domain filtering, non-stationary noise

摘要： 与单域图像去噪方式不同，双域图像去噪方式将工作于不同域的两种去噪算法有效结合到一起，能在保持图像细节的同时抑制图像噪声。但是双域图像去噪需已知图像噪声方差，在实践应用中很难精确确定。为此针对图像含有的非平稳噪声，基于双侧滤波与短时傅里叶变换（short-time Fourier transform，STFT）的双域滤波方法，开展三方面工作：（1）用含噪图像与双域滤波去噪图像间的差异图像估计噪声图像；（2）基于双侧滤波权系数估计非平稳噪声方差；（3）改进双域滤波，使之适应非平稳噪声情况。实验结果及相关定量分析表明，与相应的双域滤波相比改进算法可以更有效地保护图像细节并抑制图像噪声。

关键词: 图像去噪, 双侧滤波, 短时傅里叶变换, 双域滤波, 非平稳噪声