Abstract: Based on combining shock filter with anisotropic diffusion filter to preprocess the noisy images, the edge-enhanced image is got while smoothing out the noise in this paper. By using the edge detection filters to choose the partial adaptive parameters based on the preprocessed images, a local adaptive total variation regularization model is proposed for the image denoising problem. The proposed model can keep the balance between noises smoothing and edges preserving adaptively. Furthermore, a first-order primal-dual algorithm is used to solve the proposed the model. The numerical results show that the proposed model and algorithm can effectively smooth out the noise, at the same time it can greatly keep the details and texture features of images. Then the proposed model can improve the objective evaluation standard and subjective visual effect.

Key words: image denoising, adaptive weighted function, total variation model, primal-dual algorithm

摘要： 综合利用冲击滤波器和非线性各向异性扩散滤波器对含噪图像做预处理，然后基于边缘检测函数建立反映图像局部特征的自适应权函数，构建能同时兼顾图像平滑去噪与边缘保留的局部自适应性的全变分模型，并建议用本原对偶算法快速求解。实验结果表明，同传统的全变分图像去噪模型相比，该局部自适应全变分模型在消除噪声的同时能很好地保持图像的边缘轮廓和纹理等细节特征，得到的复原图像在客观评价标准和主观视觉效果方面均有所提高。

关键词: 图像去噪, 自适应权函数, 全变分模型, 本原对偶算法