I'm sorry if I'm being vague here, but where has this topic headed?
I still don't quite understand the principle you are explaining here, on the one hand you are going on about an image exploit trough HTML markup, and then you are redirecting your opinion to raw output? I just do not see the relevance here.
I, and I'm quite sure 98% of web developers don't filter their output, we KNOW what it is, we designed the application and the manner of output.
You only filter the input of users, yes, granted that there is allot of aspects within the term input, but it can be covered by merely a line of scripting - for each input, or actually not.
Headers should be filtered, yes.
Additionally, whenever a user enters anything to your website, it should be check for consistency.
I personally don't like to filter the input from users, I try to minimize the amount of data the user has to enter, it makes everything easier for me and the potential client.
Validation might be a bit more risky and hard to correct the perfect pattern, but when you know what you are doing, just doing regex validation on strings is much more convenient.
Now on to the topic again, can you please explain your scenario about the images in more - extensive, if possible - detail?