Test this in a new question/answer:
![*google!*][1]
[1]: http://www.google.com.ar/images/nav_logo6.png
You will see that it fails because it will parse the * as italics. It should be ignored within links and images.
Trying it now:
![google!][1]
[1]: http://www.google.com.ar/images/nav_logo6.png
This isn't technically a markdown problem. The HTML sanitizer is getting it.
It ends up like this:
<p>
<img src="http://www.google.com.ar/images/nav_logo6.png"
alt="<em>google!</em>" />
</p>
The sanitizer kills the whole thing because of embedded tags. Which I agree with. Well, I wrote the code, so I'm biased, but you know what I mean.
Just escape the first star.
![\*google!*][1] [1]: https://www.google.com.ar/images/nav_logo6.png
This is a problem with including images generated by one of the many LaTeX servers. For these, the alt text should be the exact LaTeX code used to produce the image: it makes it accessible to those who use screen readers to read the alt text, and it makes it possible to simply cut-and-paste the formula without having to go through and undo all the Markdown escaping. Given how often the underscore is used in LaTeX formulae, it would be extremely useful to have this fixed in the markdown implementation used.
This has been fixed in the PHP markdown port (even the version without all the extras) and, I believe, in the maruku ruby variant.
