Interesting idea. As niralsoni above me mentioned, there is existing software that "interprets" drawn images, so your idea is quite feasible. I don't have much to offer, but I can venture some criticism.. Not to dampen your innovation.
Some of the problems that come to mind:
Like speech recognition, software that deals with interpreting 'more human' sources of data tends to be quite.. err.. fallible. Programs like these have made big leaps forward in recent years, but still have this reputation. All I'm saying by this is, that even with an ideal result to such a project I wouldn't expect the program to always work as intended. Converting a box model might not be so problematic, but things like margin vs padding, border vs outline, background vs new element, or many other things would be tricky. Not only to get correct initially, but also to allow a non-savvy user to fix up without a lesson or two in HTML/CSS first (which this idea is seemingly meant to be dodging in the first place).
From my point of view at least, a pretty page is the least of my worries when creating a website. Actually, I'd say you could learn enough HTML and CSS in a months' free-time — if you really wanted to — to whack up a cool page to your own specifications. Anything but the most simplistic web-pages these days requires a dynamic scripting language behind it, and one of the main problems with creating a page right off an image is just that; It would be a reflection of an image. The web is dominated now by smart pages with adaptable content, and I dare say most people would get closer to their goals with a Wordpress theme that already incorporates scripts.
The workings of HTML also depends a lot on it's environment. Even if the generator could produce desired code, the user would still need to understand the structure of their web-server enough to correctly reference their resources, and place their code. Images are separate files, and ideally so are styles. Linking to these in the HTML generally begs at least basic knowledge. I'd just like to point out that there would be numerous pitfalls involved with letting a coding-illiterate person jump into web-page design.