基于PDTDFB变换域各向异性双变量模型和非局部均值滤波的图像去噪方法

Image Denoising Using Anisotropic Bivariate Laplacian Model in PDTDFB Domain and Nonlocal Means Filter

为了有效去除图像噪声并保留更多图像细节信息,提出一种结合PDTDFB变换域各项异性双变量拉普拉斯模型和非局部均值滤波的自适应图像去噪算法。首先分析了PDTDFB变换系数的分布特点,使用各项异性双变量拉普拉斯模型作为其父子系数相关性的先验分布,在贝叶斯去噪框架下推导出闭式形式的各项异性双变量阈值函数,然后对估计的变换系数进行逆PDTDFB变换得到初步去噪图像,最后使用非局部均值滤波对其进行平滑处理。实验结果显示:本文所提算法去噪效果明显,与一些经典算法相比,本文方法在主客观上皆取得了有竞争力的结果。

This paper presents a new image denoising algorithm to effectively remove noise while retaining the important marginal and detail information of images.It is based on a combination of the anisotropic bivariate Laplacian model in the pyramidal dual-tree directional filter bank transform(PDTDFB)domain and a non-local means filter(NLM)in the spatial domain.First,the distribution characteristics of the PDTDFB transform coefficients are analyzed.Then,the coefficients are modeled as an anisotropic bivariate Laplacian model considering the child-parent statistical dependency between the PDTDFB coefficients.With this statistical model,a closed-form anisotropic bivariate shrinkage function is derived in the framework of the Bayesian theory.Second,an inverse PDTDFB transform is performed to obtain the initial denoised image.Finally,NLM is used to smooth the initial denoised image.The simulation results show that the proposed method provides promising results and is competitive with the classical denoising results reported in the literature both in terms of peak signal-to-noise ratio and visual quality.