光学技术, 2018, 44 (2): 194, 网络出版: 2018-05-01
全变分引导的双边滤波图像去噪方法
Total variation guided bilateral filtering for image denoising
图像去噪 双边滤波 全变分模型 Split Bregman迭代方法 结构保持能力 image denoising bilateral filtering total variation model split bregman structure preserve capability
摘要
双边滤波是一种结合了图像空间邻近度和像素值相似度的非线性滤波方法。当噪声水平较小时, 使用输入的待去噪图像来引导双边滤波核函数权重的计算较为可行, 但当噪声水平变大时, 噪声图像的结构信息被严重破坏, 影响了双边滤波器中值域核函数权值的准确计算。提出了全变分引导的双边滤波图像去噪方法, 给出其Split Bregman快速算法。利用全变分模型对噪声图像进行光滑, 得到一幅包含清晰结构信息的图像, 将图像作为双边滤波的引导图像进行计算, 对上述过程进行迭代处理以提高算法的稳定性。实验表明, 提出得算法不仅在主观视觉效果上优于双边滤波, 客观评价指标PSNR值也较双边滤波提高了1.5dB左右;在较好去除噪声的同时较好地保持了边缘结构。
Abstract
Bilateral filtering is a nonlinear filtering method which combined the spatial proximity and the pixel value similarity.When the noise level is low, it is feasible to use the input noisy image to guide the computation of the bilateral filtering kernel function. However, when the noise level is high, the structural information of the noisy image is severely damaged, affectes the accurate calculation of the range kernel function. In order to improve the denoising effect of bilateral filtering, a total variation guided bilateral filtering method is proposed, and its fast algorithm is also given. The noisy image is smoothed by the total variation model and an image with good structure information is obtained. Then it serves as the guided image of the bilateral filtering. To improve the robustness of the algorithm, an iterative processing is used. The experimental results indicate that the proposed method is significant both visually and in terms of PSNR. It can effectively remove noises while preserving more structure and edge details.
芦碧波, 王乐蓉. 全变分引导的双边滤波图像去噪方法[J]. 光学技术, 2018, 44(2): 194. LU Bibo, WANG Lerong. Total variation guided bilateral filtering for image denoising[J]. Optical Technique, 2018, 44(2): 194.