摘要
传统去噪算法不能在尽量滤除噪声的同时很好地保持原始图像信息。针对这种情况,提出基于鲁棒主成分分析的自适应视频去噪算法。首先根据视频数据的低秩性和噪声的稀疏性,利用加速近端梯度方法重建出原始视频的低秩部分和稀疏部分,实现噪声的初步分离;其次利用自适应中值滤波器进行预滤波处理,提高块匹配精度,进一步去除视频噪声;最后引入自适应奇异值阈值法,增强图像细节边缘信息,降低迭代优化算法的时间复杂度。实验结果表明,该方法不仅能极大程度地恢复出原始视频序列,还能自适应地去除干扰噪声。不论从客观指标PSNR值还是从主观视觉,该方法与传统去噪方法相比都具有很大的优势。
Traditional denoising algorithm cannot well reserve primitive image information while filtering the noise as much as possible. In light of this situation, the paper presents an RPCA-based adaptive video denoising algorithm. First, according to the low-rank property of video data and the sparsity of noise, it utilises the accelerated proximal gradient approach to reconstruct the low-rank component and sparse component of original video, and realises the initial separation of the noise. Then, it uses adaptive median filter to make pre-processing of filtration to improve block matching accuracy, and further removes video noise. Finally, it introduces adaptive singular-value threshold method to enhance the detailed edge information of image, reduces the time complexity of iterative optimisation algorithm. It is demonstrated by experimental result that the proposed algorithm can restore original video sequence to a great deal extent, besides it can also adaptively remove interference noise. The algorithm has significant advantage no matter in objective quantitative indicator PSNR or subjective vision quality compared with traditional denoising algorithms.
出处
《计算机应用与软件》
CSCD
2016年第9期215-220,共6页
Computer Applications and Software
关键词
视频去噪
低秩性
鲁棒主成分分析
自适应奇异值阈值
Video denoising Low-rank property Robust principal component analysis （RPCA） Adaptive singular-value threshold
