Tone-Mapped Image Quality Database (TMIQD) contains 10 natural and 10 computer generated High Dynamic Range (HDR) images, tone-mapped using various operators. The primary purpose of the database is training and testing of objective image quality metrics, either specialized on tone-mapping or general quality assessment.
MATLAB scripts for calculating scores of new metrics and analyzing their performance are included as well to enable reproducible research results on a common test bed.
10 natural and 10 synthetic HDR images in .hdr format containing physical luminance values.
9 versions of each reference, i.e. 180 tone-mapped images.
5 tone-mapping operators (TMOs) have been used to create the images – Drago, iCAM06, Mantiuk, Mai, and Simple (linear mapping with clipping and inverse gamma correction). Two versions with different settings of parameters have been created for each TMO, except for Mai which does not allow adjustment of parameters. Refer to the paper for more information about the tone-mapping process.
The subjective data has been collected using Adaptive Square Design Pair Comparison methodology in two scenarios – with and without the HDR reference available (see the images on the right).
1A – natural content, setup with HDR reference
1B – natural content, setup without HDR reference
2A – synthetic content, setup with HDR reference
2B – syntetic content, setup without HDR reference
The analysis of the results showed that, in case of natural images, the outcome can be significantly influenced by the presence of the reference. We therefore recommend to select the particular part(s) of the database, relevant to the application area of the tested metric. For example:
Full-reference metric for natural tone-mapped images: Use the results from the scenario 1A
No-reference metric for natural images: Use the results from the scenario 1B
General purpose full-reference metric for tone-mapped images: Use the results from the scenarios 1A and 2A
General purpose no-reference metric: Use the results from the scenarios 1B and 2B
For more information on how to select the appropriate part(s) for the analysis, refer to the “analyze_metrics_performance.m” function provided with the database.
If you use any part of the provided dataset in your research, we kindly ask you to cite the paper:
L. Krasula, M. Narwaria, K. Fliegel, and P. Le Callet, Preference of Experience in Image Tone-Mapping: Dataset and Framework for Objective Measures Comparison, IEEE Journal of Selected Topics in Signal Processing, 2017.
If you have any questions regarding the database, code, the paper, or you want to report a malfunction of the software, feel free to contact us by email to L.Krasula@gmail.com.