I'm not literally blind, but I'm legally blind, with corrected acuity on the order of 20/200. My condition could be more informatively described as "low vision". So I'm just as dependent on my vision as a normally sighted person, but I can answer your question to some extent.

The most important rule is that text should be text. That is, [don't use images of text when you could use real text](https://stats.meta.stackexchange.com/a/3176), or at least accompany the image with a transcript or alt-text. For mathematical notation, use MathJax. The advantage here for the visually impaired is that real text, unlike images, can be enlarged without loss of quality, reflowed (trying to read a paragraph that requires horizontal scrolling will drive sane men mad), and read with text-to-speech or a refreshable Braille display. Unrelatedly to accessibility, but perhaps more importantly, text is more useful for software: it can be searched, indexed, and translated, it can transmitted more easily through media without image support, and it takes up less disk space.

Images are still going to be important in data analysis because of data graphics, like plots. The text-should-be-text rule doesn't get you far here because tick labels etc. aren't very useful to somebody who can't see the image. But I can still make suggestions to help make images legible to the visually impaired. Aim for high contrast (e.g., use black text on white rather than gray text on white or colored text). Avoid color combinations that will run afoul of the most common types of color-blindness. Avoid unwanted moiré effects from tightly packed parallel stripes. Edward Tufte's *Visual Display of Quantitative Information* advocates a sparse visual style that in general makes plots more accessible to the visually impaired, although it doesn't pursue accessibility deliberately.