基于各向异性扩散偏微分方程的图像去模糊

利用图像结构张量导出的各向异性扩散滤波,具有平滑噪声的同时保持细节的特点,提出基于结构张量的能量最小化去卷积正则化模型,并对由此导出的偏微分方程应用于灰度图像和向量值图像去模糊作了分析。灰度图像的各向异性扩散滤波可以由梯度平滑的结构张量实现,相应的偏微分方程取决于平滑结构张量决定的惩罚函数。与其它非线性扩散滤波去模糊的方法比较结果证实所提方法在信噪比和视觉质量上都具有更好的效果。

Image Deblurring with Anisotropic Diffusion Partial Differential Equations

Anisotropic diffusion filtering with image structure tensor bear characteristics which the noises are removed and discon- tinuities are reserved.An image model of regularization deconvolution based on structure tensor is proposed.The corresponding partial differential equations applied to scalar image and vector-valued image have been analyzed.Anisotropic diffusion for scalar image can be performed by structure tensor with presmoothed gradient,but its corresponding partial differential equation comes from a variational for- mu]ation depending on penalty function on smoothed structure tensor.Experiment results show this method improvement in signal-to- noise and visual effect over some existing debluring methods based on nonlinear diffusion filtering.