I came to a conclusion that the surrounding text and the page layout were the two main factors...
I agree, based on many test searches and occasional semi-systematic observation over the years, but I'm sure there are some other strong factors as well. For certain kinds of images, image file names and the alt attribute may play a part. Depends on how hard up Google is for clues. Worth noting that "alt" text has been badly misused, and its ranking value for images probably comes and goes.
I should add that I've seen Google get an image wrong in one of the upper right hand Knowledge Graph entity results, for a local cultural/ tourist attraction. I found that the error apparently stemmed from a mislabeled image in a fancy web lifestyle magazine, which had an article all about some other place in town with a similar name, but incorrectly illustrated with this image. The image itself was perhaps the best photographed shot of the place in Image Search, so it received a lot of attention there. I assume that the misidentification (not uncommon for this particular place), coupled with the popularity of the website that published it, led to the KG problems... but the mistake was fairly common in Image Search.
For me, the search served as a navigational search that I used frequently enough that it was easy for me to watch the KG results. I let the image sit for a couple of months in KG before I reported the mistake... and it took Google another month to correct it in the KG.
I hadn't looked at it in Image Search for a long time, where I now see that what appears to be the same image is still ranking for the wrong term (coincidentally, I'd bet), but it's now ranking for a different site, which also misidentifies it. The image is appearing on both sites, but only one copy is ranking.
Getting back to the creation date, the file date is changed when I download the image, so it's basically not possible to observe how Google might see that date.
I'm thinking that the first instance I saw of the image had to be older than the current one, though, as the current ranking copy of the image has dimensions and file size quite a bit smaller than the first, so may have been downloaded and reduced. I'm not sure that all this tells us much, except that there are a lot of variables.