基于主成分分析的分块视频噪声估计

Block-based Video Noise Estimation Algorithm via Principal Component Analysis

噪声估计在视频去噪领域具有重要的研究意义.实际生活中的噪声都是未知的,然而现存的视频去噪算法通常都假定视频的噪声水平是已知的,本文提出一种基于主成分分析(Principal component analysis,PCA)的分块视频噪声估计算法.首先,基于帧间进行块匹配寻找相似块,得到差分图像以消除视频运动的影响;其次,将正态分布函数作为阈值函数简化噪声估计算法模型;最后,设置明确迭代指标使得估计的结果更加精确,且降低了计算复杂度.主观视觉效果和客观指标对比表明,本文提出的基于主成分分析的分块视频噪声估计算法比其他优秀的噪声估计算法误差小同时鲁棒性高,能准确地估计视频噪声.

Noise estimation is an important issue in video denoising applications. However, in practice the noise level is unknown in most cases, but most existing denoising algorithms simply assume the noise level is known beforehand.In this paper, we propose a block-based video noise estimation algorithm via the principal component analysis（PCA）.Firstly, similar blocks are searched by block matching between frames, and the difference image is obtained to eliminate the influence of video motion. Secondly, a thresholding function of normal distribution is used to simplify the model of noise estimation. Finally, setting clear iterative metrics makes the estimation results more accurate and reduces the computational complexity. Subjective and objective comparisons show that, compared with other state-of-art algorithms,the noise estimation of the proposed video denoising algorithm is robust against small errors and achieves outstanding denoising effect.