基于Contourlet视觉模型和奇异值分解的图像质量评价

Image quality assessment based on human visual model and singular value decomposition in Contourlet domain

针对基于像素值的图像质量评价方法忽视图像结构信息和需要完全参考图像等问题,提出了一种基于Contourlet域奇异值分解CW-SVD,部分参考图像质量评价方法(contourlet weighted singular value decomposition)。在Contourlet域中,利用奇异值向量对图像结构的表征能力,结合人眼视觉敏感性确定每个子带的视觉权重,得到每个子带的评价测度,再综合得出图像的最终评价指标。实验表明,该方法应用于4种类型的降质图像质量评价时,比峰值信噪比(PSNR、MSSIM)等算法具有更好的稳定性和更好的主客观评价一致性。

Due to the limitations of traditional image quality assessment methods including lacking the image structure information and needing a complete reference image,a new image quality assessment method based on Contourlet weighted singular value decomposition（CW-SVD） with partial reference image is presented in this paper.In Contourlet domain,the human visual sensibility（HVS） is considered to determine the weight of each subband by the image characterization from singular value vector.Finally,an image quality metric is determined by overall consideration of all the subbands.The results show that compared with the peak signal-to-noise ratio（PSNR）,mean structure similitary（MSSIM） algorithms,the presented method has a great improvement in both the consistency between subjective and objective evaluations and the stability,when it is applied to four distortion types.