计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (27): 191-193.DOI: 10.3778/j.issn.1002-8331.2008.27.061

• 图形、图像、模式识别 • 上一篇    下一篇

一种新的JPEG图像无参考客观质量评价方法

黄继风   

  1. 上海师范大学 计算机科学与技术系,上海 200234
  • 收稿日期:2007-11-13 修回日期:2008-02-25 出版日期:2008-09-21 发布日期:2008-09-21
  • 通讯作者: 黄继风

New method of no-reference objectively assessing JPEG image quality

HUANG Ji-feng   

  1. Department of Computer Science and Technology,Shanghai Normal University,Shanghai 200234,China
  • Received:2007-11-13 Revised:2008-02-25 Online:2008-09-21 Published:2008-09-21
  • Contact: HUANG Ji-feng

摘要: Parsval’s原理指出,图像压缩前、后像素值的均方误差值等同与图像压缩前、后DCT系数的均方误差值,根据这一思想设计了JPEG图像无参考客观质量评价算法。假设图像的DCT系数是Laplacian分布,首先根据JPEG图像的DCT系数估计出图像压缩前的DCT系数分布参数,并由此估算出图像压缩前、后DCT系数的均方误差值。实验结果表明,该方法所用指标能准确地描述JPEG图像的质量且与空间域均方误差指标相符。

关键词: 图像质量评价, 参数估计, JPEG图像, Laplacian分布

Abstract: According to Parseval’s theorem,the MSE in the pixel domain is equivalent to the mean square quantization error in the DCT domain because DCT is a normalized orthogonal transformation.Based on this idea,a new method of no-reference objectively assessing image quality is presented.The DCT coefficients of the ac component can be approximated as a Laplacian distribution.Firstly,estimating the Laplacian parameter from the observed dequantized DCT coefficients,amplitude distributions of DCT coefficients of the source are given,the mean square quantization error in the DCT domain is obtained.Experimental results have shown,it is in good accordance with the image quality,and in accordance with the MSE in the pixel domain.

Key words: assessment of image quality, parameter estimation, JPEG, Laplacian distribution