Konstanz Fine-Grained IQA​ Dataset

Pair comparisons (PC) of distorted images, followed by Thurstonian reconstruction of scale values, have been used in the literature for fine-grained image quality assessment. We introduce boosting techniques embedded in more general triplet comparisons (TC) that further increase the sensitivity of PCs. Boosting amplifies the artifacts of distorted images, enlarges their visual representation by zooming, increases the visibility of the distortions by a flickering effect, or combines some of the above.

We created the Konstanz Fine-Grained IQA dataset (KonFiG-IQA, Parts A and B), a subjectively annotated image quality dataset with 10 source images processed using 7 distortion types (color diffusion, jitter, high sharpen, JPEG-2000 compression, lens blur, motion blur, multiplicative noise) at 12 or even 30 levels, evenly distributed over a span of 3 JND units. The KonFiG-IQA dataset also contains a large number of subjective responses to triplet comparisons and DCR ratings obtained via crowdsourcing.

Images

• Part A contains 10 source images and corresponding distorted versions (7 distortion types at 12 levels, spaced at 0.25 JND), resulting in 840 distorted images in total.
• Part B provides the distorted images for motion blur with 30 levels of distortion, spaced at 0.1 JND.

Ratings

• Subjective responses to triplet comparisons (Part A and Part B), each with a record [source_id, distortion_type, distortion_level, rating, time_used, worker_id]
• DCR ratings for Part A, each with a record [source_id, distortion_type, distortion_level, rating, time_used, worker_id]

MATLAB Code

• Adding 7 types of distortions to each of the reference image
• Boosting the distortions with respect to a reference image by artefact amplification, zooming, and flicker.
• Reconstructing JND scales from triplet comparisons

Cite us

KonFiG-IQA is freely available to the research community. If you use our database in your research, please cite the following reference:

@ARTICLE{2021arXiv210800201M,
       author = {{Men}, Hui and {Lin}, Hanhe and {Jenadeleh}, Mohsen and {Saupe}, Dietmar},
        title = "{Subjective Image Quality Assessment with Boosted Triplet Comparisons}",
      journal = {arXiv e-prints},
     keywords = {Electrical Engineering and Systems Science - Image and Video Processing, Computer Science - Computer Vision and Pattern Recognition},
         year = 2021,
        month = jul,
          eid = {arXiv:2108.00201},
        pages = {arXiv:2108.00201},
archivePrefix = {arXiv},
       eprint = {2108.00201},
 primaryClass = {eess.IV},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2021arXiv210800201M},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}
​

Downloads

KonFiG-IQA (.zip)

422 MB download

The .zip archive contains all the material, including images, codes for introducing distortions and triplet comparison reconstructions, and subjective responses:

• images of Part A and Part B
• the MATLAB codes for adding distortions and triplet comparison reconstruction
• the subjective responses obtained from the crowdsourcing experiment are provided