Talk:Structural similarity index measure

Float not decimal
"The resultant SSIM index is a decimal value between -1 and 1"
 * What is this supposed to mean; isn't decimal the default ? I think this is probably supposed to say floating point value. — Preceding unsigned comment added by 137.222.185.228 (talk) 16:18, 28 May 2014 (UTC)
 * Agree, but you need source for that, also look into https://trac.ffmpeg.org/ticket/7825 2A00:1370:812C:91A8:91C8:6C42:6022:921E (talk) 07:31, 27 March 2020 (UTC)

Biased edits
There are some rather biased edits by what I believe are people who worked on or are associated with SSIM. It would be better if this article was backed up by references – at the moment it still seems like a product advertisement page with a lot of unreferenced claims regarding SSIM's popularity. Soulhack (talk) 15:05, 22 May 2016 (UTC)


 * I keep coming back to this article, correcting biased edits that are introduced over time, and removing uncited, subjective claims. I'd appreciate if all edits made to this page could properly cite peer-reviewed sources. Slhck (talk) 09:11, 15 July 2017 (UTC)


 * Based on the 1st (number of references are plenty now) and 8th point of https://en.wikipedia.org/wiki/Help:Maintenance_template_removal#When_to_remove I consider this problem adequately solved. Gerymate (talk) 15:55, 9 Aug 2022 (CET).

Correction in sentence
In the sentence: "Setting the weights $$\alpha,\beta,\gamma$$ to 1, the formula can be reduced to the form shown at the top of this section.", $$\alpha$$ and $$\beta$$ are indeed 1, but $$\gamma$$ must be equal to 0 to equal the formula above.

Alternatively, the formula above could be modified to include the term $$s(x,y)=\frac{\sigma_{xy} + c_3}{\sigma_x \sigma_y + c_3}$$.

Resulting in: $$\hbox{SSIM}(x,y) = \frac{(2\mu_x\mu_y + c_1)(2\sigma_{xy} + c_2)(\sigma_{xy} + c_3)}{(\mu_x^2 + \mu_y^2 + c_1)(\sigma_x^2 + \sigma_y^2 + c_2)(\sigma_x \sigma_y + c_3)}$$ anon (talk) 20:17, 09 January 2017 (GMT)


 * In the original paper, it is written that way. Note that c3 is dependent on the other coefficients. Slhck (talk) 15:02, 10 January 2017 (UTC)
 * There was a mistake here https://en.wikipedia.org/w/index.php?title=Structural_similarity&diff=943549503&oldid=941593158 now everything is right. 2A00:1370:812C:91A8:2DBF:96:3D27:97F8 (talk) 08:44, 27 March 2020 (UTC)

SSIM non-negativity
The 'Mathematical Properties' section claims that SSIM exhibits non-negativity, however I don't see any evidence of this after skimming the cited paper. This also contradicts the statement in the next section: "The resultant SSIM index is a decimal value between -1 and 1, ...". This paper claims SSIM can take negative values (page 4, subnote 1) as does this tensorflow api doc. User:Kaizeng045 Evidlo (talk) 01:50, 26 April 2019 (UTC)

No references.
I deleted a large section under the applications section, which was a text just talking about the popularity of the algorithm without any clear references to back it up. Ironically the text said that the success of SSIM is evident from the large number of citations, without giving any citations for the citations or examples of well cited work which has cited the text.

Also the text talked about how SSIM was paramount for "the team winning a grammy", but this seems to be out of context. It's not said who they are, and the whole thing seemed mal placé.

However, I might just be a n00b, so please feel free to oppose me on this point and argue as to why the text should stay :) — Preceding unsigned comment added by Miketehbike (talk • contribs) 15:23, 18 March 2019 (UTC)


 * Good call. I fully agree with these actions. Slhck (talk) 16:02, 27 March 2019 (UTC)

COI notice
I have always been worried about the state of this article, which some authors treat like a place for advertising rather than neutral, factual, and unbiased statements. Many edits are performed by the algorithm's authors, or people closely associated with them. They seek to frame SSIM and its variants as the best and only quality assessment methods available. This includes repetitions of uncited claims and unneeded hyperbole, as well as in-depth descriptions of commercial implementations that have not been independently verified and aren't even available for testing. I therefore removed or modified a large portion of recent edits and added a Conflict of Interest template. My hope is that the authors in question will see this notice and act accordingly. Slhck (talk) 16:01, 27 March 2019 (UTC)