The SSIM index is not a metric and it is badly evaluate the similarity of images
https://doi.org/10.21122/2309-4923-2019-2-12-17
Abstract
The article explored some properties of a very popular feature of image structural similarity, called the SSIM index. According to https://scholar.google.com, the article [3], where it was first described, has made more than 20,800 citations during the last 14 years. This indicator is actively used by the scientific community in imaging research. It acquired the status of an unofficial international standard for assessing image quality in the presence of a template, often referred to as the image quality metric. This article debunks some of the myths that have arisen around this index. A theorem is proved which states that the SSIM index and any of its linear transformations are not metric functions. In many publications and in the Matlab application software package in the description of the SSIM function, it is said that the SSIM index is used to measure the image quality. However, this index, as well as any comparison function with a reference image (such as full-reference), in principle, cannot assess the quality of the analyzed images. They estimate only a certain degree of similarity between the template image and its distorted copy. The article also shows that the SSIM index cannot always correctly determine the similarity of images of the same scene, while the Pearson linear correlation coefficient makes it much faster and more accurate.
About the Author
V. V. Starovoitov
United Institute of Informatics Problems, National Academy of Sciences of Belarus
Belarus
Starovoitov Valery, Doctor of Sciences and professor, Principal research fellow
Belarus
Starovoitov Valery, Doctor of Sciences and professor, Principal research fellow
References
1. Pedersen, M., Hardeberg, J. Y. Full-reference image quality metrics: Classification and evaluation // Foundations and Trends® in Computer Graphics and Vision, 2012), V. 7, № 1. – pp. 1–80.
2. Lin W., Kuo C. C. J. Perceptual visual quality metrics: A survey // Journal of Visual Communication and Image Representation. – 2011. – Т. 22. – №. 4. – pp. 297–312.
3. Wang, Z., Bovik, A. C. Modern image quality assessment // Synthesis Lectures on Image, Video, and Multimedia Processing, 2006, V. 2. – № 1. – pp. 1–156.
4. Wang, Z., Bovik, A. C., Sheikh, H. R., and Simoncelli, E. P. Image quality assessment: from error visibility to structural similarity // IEEE Trans. Image Processing, 2004, V. 13. – № 4. – pp. 600–612.
5. Chandler, D. M. Seven Challenges in Image Quality Assessment: Past, Present, and Future Research’, ISRN Signal Processing, 2013, V. 2013. – C. 1–53.
6. Renieblas G. P. et al. Structural similarity index family for image quality assessment in radiological images // Journal of Medical Imaging. – 2017. – Т. 4. – №. 3. – С. 035501.
7. Sheikh H. R., Sabir M. F., Bovik A. C. A statistical evaluation of recent full reference image quality assessment algorithms // IEEE Transactions on image processing. – 2006. – Т. 15. – №. 11. – pp. 3440–3451.
8. Dosselmann R., Yang X. D. A comprehensive assessment of the structural similarity index // Signal, Image and Video Processing. – 2011. – Т. 5. – №. 1. – pp. 81–91.
9. Hore A., Ziou D. Image quality metrics: PSNR vs. SSIM // Proc. Of the 20-th Int, Conf. on Pattern Recognition, 2010. – pp. 2366–2369.
10. Pambrun J. F., Noumeir R. Limitations of the SSIM quality metric in the context of diagnostic imaging // Proc. of the IEEE International Conference on Image Processing, 2015. – pp. 2960–2963.
11. Старовойтов В. В. Уточнение индекса SSIM структурного сходства изображений // Информатика. – 2018. – Т. 15. – №. 3. – С. 41–55.
12. Ponomarenko, N, Jin, L, Ieremeiev, O, Lukin et al. Image database TID2013: Peculiarities, results and perspectives // Signal Processing: Image Communication, V. 30. C. 57–77.
13. Gu, K., Zhai, G., Lin, W., Liu, M. The analysis of image contrast: From quality assessment to automatic enhancement’ // IEEE Transaction on Cybernetics, 2016. – V. 46. – № 1. – С. 284–297.
Review
For citations:
Starovoitov V.V. The SSIM index is not a metric and it is badly evaluate the similarity of images. «System analysis and applied information science». 2019;(2):12-17. (In Russ.) https://doi.org/10.21122/2309-4923-2019-2-12-17
Views:
833
Email this article 