This is really tricky, because regular expressions are mostly about matching something that is there. With look-around trickery, you can do things like 'find A that is not preceded/followed by B', etc. But I think the most pragmatic solution for you wouldn't be that.
My proposal relies a little bit on your existing code not doing too crazy things, and you might have to fine-tune it, but I think it's a good shot, if you really want to use a RegEx-search for your problem.
So what I suggest would be to find all img tags, that can (but don't need to) have all valid attributes for an img-element. Whether that is an approach you can work with is for you to decide.
Proposal:
/<img\s*((src|align|border|height|hspace|ismap|longdesc|usemap|vspace|width|class|dir|lang|style|title|id)="[^"]"\s*)*\s*\/?>/
The current limitations are:
- It expects your attribute values to be delimited by double quotes,
- It doesn't take into account possible inline on*Event attributes,
- It doesn't find img elements with 'illegal' attributes.