Abstract: For Total Variation（TV） model vulnerable to gradient and regularly failing to capture image details, this paper proposes an adaptive total variation denoising algorithm based on curvature differential. On the basis of preprocessing the noisy images with both nonlinear anisotropic diffusion filter and shock filter, the proposed algorithm can make a tradeoff between edge-preserving and noise-smoothing by adaptively adjusting the weights of the regular item and the fidelity item. The experimental results show that compared with the existing denoising algorithms, the proposed new algorithm can realize at least 14% increasement of peak signal to noise ratio, and at least 43% reduction of normalized mean square error under impulsive noise in different intensity.

Key words: image denoising, total variation, adaptive, detail preserving

摘要： 针对整体变分（TV）修复模型易受到梯度的影响而且常常会丢失图像细节信息的缺点，提出了一种基于曲率差分的自适应全变分去噪算法。在联合非线性各向异性扩散滤波器和冲击滤波器对含噪图像做预处理的基础上，通过自适应方式调节正则项和保真项的权重系数，该算法能同时兼顾边缘保留和图像平滑去噪。仿真实验结果表明：与现有的去噪算法相比，该算法在不同强度的脉冲噪声下可以将峰值信噪比提升14%以上，同时将归一均方误差降低43%以上。

关键词: 图像去噪, 全变分模型, 自适应, 细节保护