Online Materials for RASSDL

Online Materials for RASSDL

We compare our method with 4 existing tone mapping operators (TMOs), including QiuTMO [1], DuanTMO [2], LiangTMO [3] and RanaTMO [4]. QiuTMO is a global method. DuanTMO and LiangTMO are local methods. For the above 3 competitors, we use their default parameters for testing, instead of manually adjustiing parameters for each testing image. Generally, traditional TMOs suffer from the selection of parameters. RanaTMO is a deep learning based method which constructs a deep TMO based on the best results of 13 conventional TMOs. In the comparison, for RanaTMO, we use the same training HDR images in Fairchild [5] as our offline model. In this page, we show some materials including some visual comparisons and the comparison based on mean opinion score (MOS).

1. Comparison Based on MOS

A total of 30 subjects participated in the test to acquire reliable MOS. The subjects, including 18 males and 12 females, have an age range from 18 to 35. The physical environment of the experiments is fixed as the same as our previous work in [6]. In the test, there are 6 results for each testing image. Everytime we radomly showed 2 results to subjects, and subjects were asked to choose the better one of these 2 results. This operation continues until that 6 results were rated by the subjects. For 6 results of each testing image, the “1” value represents the best one and the “6” value represents the worst one. The final MOS of 3 databases, i.e., HDR-Eye [7], Anyhere [8] and Stanford-HDRI [9], were illustrated in the following table. Smaller values of MOS indicate better qualities of tone mapping results, and vice versa.

2. Visual Comparisons

(a) QiuTMO [1]

(b) DuanTMO [2]

(c) LiangTMO [3]

(d) RanaTMO [4]

(e) RASSDL(offline)

(f) RASSDL(online)

(a) QiuTMO [1]

(b) DuanTMO [2]

(c) LiangTMO [3]

(d) RanaTMO [4]

(e) RASSDL(offline)

(f) RASSDL(online)

(a) QiuTMO [1]

(b) DuanTMO [2]

(c) LiangTMO [3]

(d) RanaTMO [4]

(e) RASSDL(offline)

(f) RASSDL(online)

(a) QiuTMO [1]

(b) DuanTMO [2]

(c) LiangTMO [3]

(d) RanaTMO [4]

(e) RASSDL(offline)

(f) RASSDL(online)

(a) QiuTMO [1]

(b) DuanTMO [2]

(c) LiangTMO [3]

(d) RanaTMO [4]

(e) RASSDL(offline)

(f) RASSDL(online)

References

[1] G. Qiu, J. Guan, J. Duan, and M. Chen, “Tone mapping for hdr image using optimization a new closed form solution,” in 18th International Conference on Pattern Recognition (ICPR’06), vol. 1, 2006, pp. 996–999.

[2] J. Duan, M. Bressan, C. Dance, and G. Qiu, “Tone-mapping high dynamic range images by novel histogram adjustment,” Pattern Recognition, vol. 43, no. 5, pp. 1847–1862, 2010.

[3] Z. Liang, J. Xu, D. Zhang, Z. Cao, and L. Zhang, “A hybrid l1-l0 layer decomposition model for tone mapping,” in The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018.

[4] A. Rana, P. Singh, G. Valenzise, F. Dufaux, N. Komodakis, and A. Smolic, “Deep tone mapping operator for high dynamic range images,” IEEE Transactions on Image Processing, vol. 29, pp. 1285–1298, 2020.

[5] M. Fairchild, “The hdr photographic survey,” 2008. [Online]. Available: http://rit-mcsl.org/fairchild//HDR.html.

[6] F. Zhou, R. Yao, B. Liu and G. Qiu, “Visual Quality Assessment for Super-Resolved Images: Database and Method,” IEEE Transactions on Image Processing, vol. 28, no. 7, pp. 3528-3541, July 2019.

[7] H. Nemoto, P. Korshunov, P. Hanhart, and T. Ebrahimi, “Visual attention in ldr and hdr images,” 2015. [Online]. Available: http://infoscience.epfl.ch/record/203873.

[8] G. Ward and A. Software, “High dynamic range image examples,” 2006. [Online]. Available: http://www.anyhere.com/gward/hdrenc/pages/originals.html.

[9] F. Xiao, J. M. DiCarlo, P. B. Catrysse, and B. A. Wandell, “High dynamic range imaging of natural scenes,” Color and Imaging Conference, vol. 2002, no. 1, pp. 337–342, 2002.

Contact

Fei Zhou: flying.zhou@163.com

Guangsen Liao: liaoguangsen2018@email.szu.edu.cn