一种基于核函数的各向异性扩散图像去噪算法

Image Denoise Model Based on Kernel Anisotropic Diffusion

分析了各向异性扩散去噪模型优缺点,针对PM模型不能有效区分噪声和边缘,提出了一种基于核函数的各向异性扩散去噪模型。在该模型中,把图像中噪声与边缘在低维空间的非线性区分关系转变为高维特征空间的线性关系,利用核函数获得高维空间的扩散函数。实验中分别与PM模型、Cattle模型比较分析,证明基于核函数的扩散模型在去除噪声的同时,更好地保留图像的信息,且峰值信噪比最高,去噪性能最优。

Through discussing the characteristics of the anisotropic diffusion model,a novel kernel anisotropic diffusion model is proposed to overcome conventional anisotropic diffusion that fails to discriminate between noise and edges. This novel model transforms the image into a feature space with high dimensionality, so the nonlinearly separable patterns of edges and noise are linearly separable in the high dimensionality, then incorporates a kernelized gradient operator to obtain diffusion function. Compare to PM model and Cattle model, numerical experiments results show the kernal diffusion model can remove the noise while preserving more image details and gain higher peak signal-noise ratio, has good performance in image denoising.