基于SVD的超分辨率重建图像质量无参考评价方法（.docVIP

基于SVD的超分辨率重建图像质量无参考评价方法( 黄慧娟，禹晶，孙卫东 （清华大学电子工程系北京 100084） （tcyzhhj@） 摘要: 图像重建质量评价是衡量超分辨率方法优劣的决定性指标，现有的图像重建质量评价方法大都需要借助同一场景的更高分辨率图像作为参考图像，但在实际超分辨率重建过程中这样的高分辨率参考图像是不存在的。考虑到人眼对相对亮度变化更为敏感，且边缘清晰度与图像主观评价质量成正比，本文利用去均值图像、梯度图像以及灰度图像，提出了一种仅依靠图像自身信息的、基于SVD的超分辨率重建图像质量无参考客观评价方法。该方法首先将图像分割为相互重叠的局部图像块，其次在各局部图像块中利用SVD分别得到低分辨率图像间以及它们与超分辨率重建图像间的子评价结果，然后将它们再整合为最终的客观评价结果。实验结果表明，该客观评价与人类主观评价结果具有很好的一致性。 关键词: 超分辨率重建，图像质量评价，奇异值分解，无参考图像 中图法分类号:TP 391.4 A No-Reference Svd-Based Image Quality Assessment Method for Super-Resolution Reconstruction Huang Huijuan，Yu Jing，Sun Weidong （Department of electrical engineering，Tsinghua University，BeiJing 100084） （tcyzhhj@） Abstract:Image quality index is one of the most critical factors to evaluating the reconstruction efficiency of super-resolution algorithm. But many existing image quality assessmentmethods need a higher resolution image as the reference image, which could not be satisfied in the practical super-resolution processing procedures.In this paper, taking into account the facts that human eyes are more sensitive to the relative luminance and edges sharpness is proportional to the subjective image quality assessment, anobjective no-reference SVD-based image quality assessmentmethod for the reconstructed super-resolution images which adopted the mean removing images, gradient images and gray images has been proposed.First of all, the image to be evaluated will be divided into several different overlapped patches. And then, sub-image quality indexes of the low resolution images and the reconstructed super-resolution image against themare calculated based on SVD in the area ofneighborhood. Finally, all of the sub-image quality indexes are pooled together into one total image quality index. Several experimental results demonstrated that this proposed objective super-resolution image quality indexis more consistent with human subjective assessment results. Keywords：Super-resolution reconstruction, Image quality assessment, Sigular Value Decomposit