基于峰度系数方差的加权自适应引导滤波

针对传统局部窗口方差提取有噪图像边缘困难的问题，本文介绍了峰度系数方差准确感知有噪图像边缘，并基于此提出一种加权自适应引导滤波器.自适应调节引导滤波中固定正则化系数，解决边缘模糊、边缘光晕的问题；基于像素的多邻域窗口边缘权重，实现聚合加权滤波，进一步抑制边缘模糊.实验结果表明:本文方法在不同噪声级别下，PSNR、SSIM均值提升率分别大于65%、78%；滤波结果图像在噪声平滑和边缘保持两方面均有更好性能. 

Weighted Adaptive Guided Filtering Based on Kurtosis Variance

To overcome the difficulty in extracting edge of noise image with traditional local window variance, a kurtosis coefficient variance was introduced to percept edges of noise image. Meanwhile, a novel weighted adaptive guiding filter (WAGIF) was proposed. Firstly, the fixed regularization parameter of traditional guiding filter was adjusted adaptively based on edge information of local window. Secondly, the weighted aggregation was achieved based on the multi-neighbor window edge weight. And then edge blurring was further suppressed. Experimental results show that the average PSNR and SSIM on different noise levels are boosted above 65% and 78% respectively using WAGIF. Filtering results show better performance in noise smoothing and edge-preserving according to state-of-the-art image filters.