基于自适应阈值参数的改进各向异性扩散模型

An Improved Anisotropic Diffusion Model Based on Adaptive Threshold Parameters

针对传统的各向异性扩散PM模型在对图像去噪时会产生阶梯效应,并且会丢失细节的问题,提出了一种改进的各向异性扩散IPM模型.该模型用梯度模、局部方差以及基于差分曲率的自适应边缘阈值参数构造了新的扩散系数,在边缘、细节、平坦区域进行不同强度的扩散.实验结果表明,IPM模型在去噪的同时不仅能有效保持边缘、细节等信息,还能克服阶梯效应,与其他基于偏微分方程的图像降噪算法相比,信噪比(SNR)、峰值信噪比(PSNR)和平均结构相似度(MSSIM)都得到了提高.

Aimed at the drawbacks of traditional anisotropic diffusion PM model,it had a staircase effect on image denoising and resulted in the loss of detail.An improved anisotropic diffusion IPM model was proposed.Gradient magnitudes,local variance and adaptive edge threshold parameter based on the difference curvature constructed a new diffusion coefficient,which diffused differently in edge,fine detail and flat areas.The experimental results show that the proposed IPM method performs efficiently in edge and fine details preservation,avoiding staircase effect while getting rid of noise.Compared with other image de-noising algorithms based on partial differential equations,the PSNR,SNR and MSSIM are improved.