Abstract: In order to improve the denoising performance of bilateral filtering, introducing local pattern of images, the gradient bilateral filtering algorithm is proposed. Using the gradient distance of the intensity values of adjacent pixels to construct a gradient similarity kernel, the weighted average of neighboring pixels in an image is performed through the functions of geometric closeness kernel and gradient similarity kernel, then filtering is achieved. In order to obtain the optimal filtering parameters, the filtering parameters are selected through an experiential learning approach. Finally the universal parameter configuration is got. The experimental results show that the new method can well preserve edges of images, and compared with the traditional denoising models, its denoising performance is also the best.

Key words: bilateral filtering, local pattern, gradient bilateral filtering, gradient similarity kernel, parameter configuration

摘要： 为了改善双边滤波的去噪性能，引入图像的局部模式，提出了梯度双边滤波算法。采用相邻像素亮度值的梯度距离来构造梯度相似度核，通过几何邻近度核函数和梯度相似度核函数来对图像邻域像素进行加权平均，从而实现滤波；为了获得最佳的滤波参数，通过经验学习的方法对滤波参数进行选择，最终得到通用的参数配置。实验结果表明，新方法能很好地保持图像的边缘，且与传统去噪模型相比，其去噪性能也是最好的。

关键词: 双边滤波, 局部模式, 梯度双边滤波, 梯度相似度核, 参数配置