Data for: SISRSet: Single Image Super-Resolution Subjective Evaluation Test and Objective Quality Assessment

This SISRSet database is established for single image super-resolution quality assessment study. For the subjective evaluation tests, there are 15 pictures chosen from Set5, Set14 and BSD100 as the ground-truth images. The corresponding LR images are obtained by Bicubic method with downscaling factors of 2, 3 and 4. There are 360 SR images generated by 8 SR algorithms with three scaling factors in total. The 8 SR algorithms include the traditional methods: Bicubic, A+, ANR, SelfExSR and the deep learning based SR methods: CSCN, SRCNN, DRCN, VDSR. We chosed the pairwise comparison method to conduct the subjective evaluation test. There are 16 participants without knowledge of the ground-truth and SR images. The setting of the viewing environment and the test condition follow the ITU-R BT.500-11 standard.

All original images and the SR images are in the file package. Their MOS values and deviation are also included. The codes of several IQA metrics are introduced in the file. Meanwhile, some experimental results are shown in the file.

本 SISRSet 数据库旨在构建用于单图像超分辨率质量评估的研究。针对主观评估测试，选自 Set5、Set14 和 BSD100 的15幅图像作为基准图像。相应的低分辨率图像通过双三次插值法，以2、3和4的降采样因子获得。总计由8种超分辨率算法生成了360幅超分辨率图像，这8种算法包括传统方法：双三次、A+、ANR、SelfExSR，以及基于深度学习的方法：CSCN、SRCNN、DRCN、VDSR。我们选取成对比较法进行主观评估测试。共有16位对基准图像和超分辨率图像均无了解的参与者。观看环境和测试条件的设置遵循ITU-R BT.500-11标准。所有原始图像和超分辨率图像均包含在文件包中。同时，还包括了它们的MOS值和偏差。文件中介绍了几种图像质量评估指标的代码。此外，文件中还展示了部分实验结果。