针对传统各向异性扩散滤波算法难以在噪声环境下有效估计边界像素，本文提出了一种热传导系数构造方法。该方法结合了各向异性扩散和各向同性扩散的优点，将每次迭代运算分解为两步：第一步采用各向同性扩散降低图像噪声，并完成热传导系数的计算；第二步运用各向异性扩散，实现真正的图像滤波。试验证明该方法能够在大尺度加性和乘性混合噪声环境下，很好的抑制噪声并保持边缘。与传统扩散滤波器对比，本文提出的方法有效提高了在各向异性扩散滤波过程中对边界像素的估计。

It is difficult for conventional anisotropic diffusion filters to estimate edge pixel in noisy condition. In this paper, a new construction method for conduction coefficient was proposed, which improved the estimation of edge pixels in diffusing iteration and showed good performance for high-level noise reducing. The experimental tests were implemented on highly noised synthetic and scene images, and results proved our method excels over traditional filters in terms of variance reduction, edge preservation, and visual comparison.