The big one is text equivalence. For every image, there should be a text that accomplishes the same thing.
All screenshots and/or images should have some alt-text appended to them. In Confluence, the way to do this is edit the doc, click the image and select properties.
The other big one is relying on pictures of text. In other words, if you have to read some words depicted in an image in order to do something, that's bad. (i.e. Don't take a screenshot of a configuration without also including the configuration settings in text.)
Alt-text should be descriptive, too. "Screenshot" doesn't cut it. Example: "This screenshot depicts a Google Chrome window displaying the website login screen." If the image has already been described in the text above or below, a brief description of what it is showing (i.e. login window) followed by "as described in text above" is sufficient. Although if it depicts something unique, that should be described in the alt text.
For additional guidance see the proposed accessibility guidelines: https://www.w3.org/TR/WCAG20/