一种新的混合噪音滤波器加速算法

A New Acceleration Algorithm for Mixed Noise Filter

混合噪音滤波器MNF是目前滤波效果最好的算法之一,然而,由于它采用非局部算法思想,所以存在较大的计算负担,针对该不足,提出一种基于均值和方差相似性的加速算法(FMNF)。该算法之关键思想是对邻域像素预分类,把两像素的均值比和方差比均在给定阈值范围内(接近于1)的邻域像素分为一类,视为相似像素,相似的像素参与滤波计算,不相似的像素被忽略,因此,减少了参与"脉冲过滤范数"计算的像素数,提高了滤波速度。仿真实验结果表明,FMNF对各种噪音类型(纯高斯噪音、纯脉冲噪音以及它们的混合噪音)去噪的视觉效果和PSNR均与MNF相当,且滤波速度均可提高15%以上。因而,FMNF算法比MNF更具有实用性。

Mixed noise filter MNF is one of the best algorithms at present. However, it uses the non-local algorithm, so there is greater calculation burden. For this shortage, a mixed noise filter acceleration algorithm （FMNF） base on mean value and variance similarity is proposed. The main idea of the algorithm is pre-classifying to neighboring pixels, the neighboring pixels of ratio of the mean value and variance within a given threshold value range （close to 1） is divided into a class, as a similar pix- els. The similar pixels take part in filtering calculation and dissimilar pixels are ignored. Therefore, it reduces the number of pixels involved in ＂impulse controlled weighted norm＂ calculation and improves the filtering speed. The simulation results show that visual effect and PSNR of denoising figure of FMNF and MNF are quite to various kinds of noise （pure Gaussian noise, pure impulse noise and its mixed noise） denoising, and the filtering speed all can be increased by more than 15~. Therefore, the FMNF algorithm is more practical than MNF algorithm.