基于Contourlet域梯度结构相似度的无参考图像质量评价

No-Reference Image Quality Assessment Based on Gradient Structural Similarity in Contourlet Domain

Contourlet变换已经成为多尺度、多方向图像处理领域的研究热点,文章将其引入图像质量评价,提出了一种基于Contourlet变换的梯度结构相似度无参考图像质量评价方法(NRCT-GSSIM),将待评价图像通过高斯低通滤波器构造参考图像,然后在Contourlet变换域中计算图像频率子带、方向子带分别加权的梯度结构相似度,实现对图像的质量评价。该方法无需标准参考图像,充分考虑了人类视觉特性。实验结果表明,NRCT-GSSIM取得了与峰值信噪比(PNSR)、结构相似度(SSIM)和梯度结构相似度(GSSIM)等有参考图像质量评价相一致的评价结果,且较之PSNR、SSIM有更高的准确性。

The contourlet transform has drawn tremendous attention in the research of multi-scale and multi-direction image processing. This paper introduces it into image quality assessment and provides a new image assessment method based on the contourlet transform and gradient structural similarity without standard images, which is called NRCT-GSSIM. This method constructs a reference image by a Gaussian low-pass filter and assesses the image quality by computing weighting gradient structural similarity either in frequency sub-bands or in direction sub-bands. The new method considers features of the human vision system and it does not need the standard reference image. Simulated results show that NRCT-GSSIM is well in accordance with quality assessment results of some full-reference methods such as the peak signal-to-noise ratio （PSNR） , structural similarity （SSIM） and gradient structural similarity （GSSIM） indexes and the computed result of NRCT-GSSIM is more accurate than the ones of PSNR and SSIM.