电子学报 ›› 2021, Vol. 49 ›› Issue (8): 1498-1506.DOI: 10.12263/DZXB.20191207

• 学术论文 • 上一篇    下一篇

基于非局部自相似序列集的一类视频图像盲去噪算法

邢远秀1,2, 李军贤1,2, 王文波1,2, 王松3   

  1. 1.武汉科技大学理学院,湖北 武汉 430081
    2.冶金工业过程系统科学湖北省重点实验室,湖北 武汉 430081
    3.南卡罗莱纳大学工程与计算学院,美国南卡罗莱纳州哥伦比亚市 29208
  • 收稿日期:2019-10-25 修回日期:2021-02-24 出版日期:2021-08-25
    • 作者简介:
    • 邢远秀(通信作者) 女,1980年8月出生于河南省南阳市.博士,武汉科技大学理学院副教授,硕士生导师.主要研究方向为图像处理、机器视觉等. E-mail:yuanxiu@126.com
      李军贤 女,1994年1月出生于河南省安阳市.现为武汉科技大学硕士研究生,主要研究方向为图像处理、计算机视觉.E-mail:1156325080@qq.com
    • 基金资助:
    • 冶金工业过程系统科学湖北省重点实验室基金 (Y202008); 国家自然科学基金 (61671338)

Blind Video Image Denoising Based on Nonlocal Self-Similarity Series Sets

XING Yuan-xiu1,2, LI Jun-xian1,2, WANG Wen-bo1,2, WANG Song3   

  1. 1.College of Science, Wuhan University of Science and Technology, Wuhan, Hubei 430081, China
    2.Hubei Province Key Laboratory of Systems Science in Metallurgical Process, Wuhan, Hubei 430081, China
    3.College of Engineering and Computing, University of South Carolina, Columbia, SC 29208, USA
  • Received:2019-10-25 Revised:2021-02-24 Online:2021-08-25 Published:2021-08-25
    • Supported by:
    • Fund for Hubei Province Key Laboratory of Systems Science in Metallurgical Process (Y202008); National Natural Science Foundation of China (61671338)

摘要:

为提高具有帧间位移平移特性的视频图像的信噪比和去噪时效,本文提出了基于非局部自相似序列集的视频图像盲去噪算法.选取与待去噪视频图像前后相邻的若干图像帧,在每一图像帧中寻找具有典型特征的图像块群,并通过在前一帧图像中查找和该图像块群具有最小差异度的块群来确定帧间的精确位移;将待去噪视频图像划分成若干图像块,根据帧间位移快速构建每个图像块的自相似序列集;随后将每个自相似序列集中的二维图像块整合成三维矩阵后进行三维变换,并对变换系数进行自适应阈值处理;再将三维逆变换后的图像块融合生成去噪图像.实验结果表明,在噪声方差未知的情况下,本文算法所得去噪视频图像具有较好的信噪比和视觉效果,并且有较高的运行效率.

关键词: 视频图像去噪, 块匹配, 自相似序列集, 三维变换, 自适应阈值

Abstract:

A blind video image denoising algorithm based on non-local self-similar series sets is proposed to improve the peak signal-to-noise ratio and denoising efficiency of video images with displacement characteristics. Image block-groups with typical features are detected in each frame, and accurate inter-frame displacement between the image and previous frame is computed based on image block-matching. The noise image is divided into several image blocks, and the self-similar series set of each image blocks is constructed quickly according to the inter-frame displacement. A 3D transform is applied to the self-similar series set, followed by an adaptive-threshold of the transform coefficients to attenuate the noise. The 3D estimate after inverse 3D transformation are aggregated to obtain the finial image. Experimental results show that the proposed algorithm has significate advantages in PSNR and visual effect in unknown noise variance, meanwhile it also has higher efficiency.

Key words: video image denoising, block-matching, self-similarity series set, 3D transform, adaptive threshold

中图分类号: