Abstract: In guided filtering, a guided image with good structure is needed. However, when noise level is high, the structure and edges in guided image are damaged. In this case, it can not provide efficient guided information and the denoising effect is not as good as expected. The classical total variational model can preserve structure and edges while producing piecewise constant image. It is a good choice for guided image. In this paper, it proposes a new image denoising approach by combining total variational model and guided filtering. The noisy image is processed firstly by total variational model and then it serves as the guided image of the guided filtering. To avoid staircase effect introducing by total variational model, an iteration strategy is used. Experimental results show that the proposed method can effectively remove noises, preserve more details and relieve the staircase effect.

Key words: image denoising, guided filtering, total variational model, edge preserving, filter kernel

摘要： 引导滤波要求具有良好结构的引导图像进行辅助滤波。当噪声较多时，引导图像中的结构、边缘等遭到破坏，无法提供有效的引导信息，严重影响去噪效果。经典的全变分模型可以得到分片常数图像，具有良好的保持边界和结构的性能，可为引导滤波提供鲁棒的引导信息。为此，结合全变分模型和引导滤波，以噪声图像作为输入，利用全变分模型处理后的图像作为引导图像，而后进行引导滤波，并对上述过程进行迭代处理，以消除全变分模型带来的阶梯效应。实验结果表明，该算法在去除噪声的同时保持更多的细节，同时减轻全变分模型的阶梯效应。

关键词: 图像去噪, 引导滤波, 全变分模型, 边界保持, 滤波核