计算机工程

• 图形图像处理 • 上一篇    下一篇

基于三维结构张量的图像模糊度评价方法

张 涛1,2,梁德群1,王新年1,刘丽娟2   

  1. (1. 大连海事大学信息科学技术学院,辽宁 大连 116026;2. 辽宁师范大学物理与电子技术学院,辽宁 大连 116029)
  • 收稿日期:2013-05-29 出版日期:2013-11-15 发布日期:2013-11-13
  • 作者简介:张 涛(1976-),女,讲师、博士研究生,主研方向:数字图像处理;梁德群,教授、博士生导师;王新年,副教授、博士;刘丽娟,副教授、硕士
  • 基金项目:
    国家部委基金资助项目;教育部博士点基金资助项目(20070151014);中央高校基本科研业务费专项基金资助项目(2012JC 038)

Image Blur Degree Assessment Method Based on 3D Structure Tensor

ZHANG Tao   1,2, LIANG De-qun     1, WANG Xin-nian    1, LIU Li-juan     2   

  1. (1. School of Information Science and Technology, Dalian Maritime University, Dalian 116026, China; 2. School of Physics and Electronics Technology, Liaoning Normal University, Dalian 116029, China)
  • Received:2013-05-29 Online:2013-11-15 Published:2013-11-13

摘要: 根据图像二维结构张量的特性和清晰图像及其模糊副本之间存在较大差异的特点,构造单幅图像的三维结构张量,其特征值大小反映了图像的局部几何结构信息,并与图像的清晰度有关,据此提出基于三维结构张量的图像模糊度评价方法。对输入图像进行不同尺度的低通滤波,得到输入图像的2个副本,并将其分成互不重叠的块。通过计算各块的三维结构张量得到模糊度参数,将各块的模糊度参数和关注度系数加权得到整幅图像的模糊度参数,以模糊度参数的指数函数形式计算图像的模糊度。实验结果表明,该方法计算出的模糊度相对于图像的模糊程度是单调的,具有良好的抗噪性,并且符合人眼视觉系统特性。

关键词: 人类视觉系统, 图像质量评价, 模糊度, 图像三维结构张, 奇异值分解, 特征值

Abstract: Theoretical and experimental studies have shown that the 2D tensor of an image can characterize the local structure and there is a high difference between a sharp image and its blurred version. Based on these studies, a novel 3D structure tensor representation of a single image is proposed, whose eigenvalues can characterize the local geometry structure and have a close relationship with image quality, and then a new blur degree assessment method is proposed. The flow chart of the proposed method blurs the input image with a weak and strong low-pass filters firstly, and divides the input image into no overlapped patches, then computes eigenvalues of the 3D structure tensor and blur measuring parameter of each block, finally computes the blur degree of an image by the exponential function of the blur measuring parameters and visual attention weights of each patch. Experimental results show that the proposed method is monotonic, robust to additive noises, and also consistent with human visual system.

Key words: human visual system, image quality assessment, blur degree, image 3D structure tensor, singular value decomposition, eigenvalue

中图分类号: