Forum Moderators: Robert Charlton & goodroi
I think that Google must have changed their algorithm to take on-page and surrounding text into major account because so many images on the web lack proper file names, alt text and/or title text, but this has just increased their false positive rates to really annoying levels. For some of my image searches, it's 1% correct, 99% incorrect.
Another problem that Google Image Search has is the duplicate indexing of hotlinked images. An Adsense-supported website has been using one of Google's own search APIs to get the top image result from their image search, and this website then inline displays the full-sized original image (hotlinks from each original website without permission) on their webpages. Despite the fact that the image and the image's web address are the same and both already exist in the Google Image Search database, Google reindexes the hotlinked images for a second time, except this time as now belonging to the hotlinker's pages!
For my own websites, I give my images proper file names and alt text (and often title text as well), and this seems to help them rank quite well, but it also helped get a number of them reindexed for this hotlinker (ugh).
But right now, it's as if I am looking for an apple -- but there's a page with a picture of a dog and also the word apple somewhere. I get a picture of a dog.
Also problematic in a similar way is a search for a person's photo, using their name in quotes. You can easily end up with some one else's photo and previously this was a lot less common.
With G my site goes in and out every few weeks and test searches that should produce two oe three relevant pages (hopefully with my pix first) go totally off topic about half way down page 1.
Also, I presume that .edu sites tend to have better PR, inherently, so instead of these unrelated images being at the bottom of the results, they are next to or within a few spots from the actual species image.