I think it is sounds doable, but I don't know how.

I have no idea how fast displays refresh. Sometimes they seem like they can work rather quickly. I think it depends more on the program that's sending information to them.

I suspect the program was having trouble reading the image. Right now, it converts the image to black and white, then scales it down to a tenth the original size. I imagine the issue is with the size of the squares I used, since the ones that showed up were 10x10 before processing, and the ones that didn't show up weren't quite so easy to handle at a tenth the size (25x25 and such).

I imagine that your help would come in handy, since it wouldn't be as useful with my contrived tests as with more complex images that would be difficult for me to figure out.
My last test was just me arranging yellow squares to form the words "I win" in braille. The output was correct, at least.
If nothing else, it might help if I could save test images at varying points of processing, and someone sighted could point out what went wrong.
The method for converting to black and white involves spreading error to surrounding pixels, which I imagine might not scale down so well.

[edit] I tried adding filters and inversion, and apparently didn't do it right, since whenever I try to use them the output is just blank space.
I tried it on a screencap of sonic 3, and the background was too bright to make much sense of the results. So being able to invert it would probably help.
Either this if statement is wrong, somehow:
if (((value < threshold)&&(!invert))||((invert)&&(value>=threshold)))
Or I should probably just remove the "error spreading" (dithering?) entirely.
[/edit]

I wouldn't mind at all.

I would hope that the size and scaling could be varied depending on what people would use it for, but it would generally be good to have the processed image no wider than 40px (or 80 at most). On height, or the size of the original image, I have no idea.

Sounds really interesting. However, I think the best tactile graphical view we can get is from a Braille printer. This would be very expensive though.
Refreshable Braille displays are able to change extremely quickly. I don't know how fast per second.
I'm wandering how all this could be useful in games, but I'm very interested to follow the project.

Ah, I'm terribly sorry CAE, I got side tracked and totally forgot about this thread.  I haven't coded anything yet.  Oops.

Do you want it to focus mostly on edge detection or shading in the entire object?  I'm assuming you want edge detection but I want to be sure.

What sources will the images be coming from, or will they be specifically created for use in this program of yours?  For example, is this thing attempting to grab images of the screen to hopefully allow someone to feel something that is on it?  Is it grabbing images of a mainstream game?  The problem I'm having is that you really can't fit much visual information into a 40x? sized black or white image.  You would never be able to represent a computer screen or frame from a mainstream game this way.

At the moment I have it load the desired image, marks the edges, and then shrinks it down into a black and white image that's 40 pixels wide and the appropriate height to maintain the original image's width to height ratio.  I'll need to add in some code to clean up the edges a bit, but in almost all cases it loses too much information to be much good.  I think the key might be to especially create graphics for this braile display, rather than attempting to shrink down existing images.  For example, mario from one of the older mainstream mario brothers games can fit into a 40x40 pixel image.  When converted to only black and white it can even still be possible to see make out his identity, but it is considerably more difficult.  If you started out with a picture of the entire mario brother's scene, mario would end up being only the size of a single pixel in that 40x40 image.  Actually he might even be smaller than a pixel meaning he wouldn't even show up.  So you'd be looking at a best case situation of a single raised dot to represent mario in the scene.  With that kind of information loss, you would have no way of distinguishing mario from an enemy, a wall, a pipe, or just about anything else.  They would all just be a single raised dot.

If you especially designed tiny black and white graphics for some braille output game, it would be very easy to represent what was going on.  As an example, I built an exact copy of the mainstream pacman game that fits into a 95x95 pixel window.  It was done as part of a project to see how small the game could be made before it was impossible to know what was going on visually.  To compare, a normal pacman sized pacman game I've made was 570x570 in size.

I don't intend to be discouraging, I just wanted to fill you in on what I'm thinking.

I can have it split it by colors if you want.  Is this going to need to capture images of the screen or an open game window?  If so, I need to code that.

I don't know of any braille fonts myself, but this latest mario example has some good possibilities.  I think that rather than making a generic all-purpose program to capture the screen, you should design it for a specific game.  If this program is generic and it processes scenes from mario brothers, you will not be able to distinguish between a pipe or the decorate cloud in the background that has no bearing on the gameplay.  If it is designed especially for mario, it can be designed to recognize bricks, pipes, and enemies so that it displays it in the easiest way to understand.

Of course I'm sure the generic all-purpose program is what is desired, so that anyone can use it on any game they wish to play.  And if we were specifically tailoring it to a particular game, we might as well just code our own version of that game which is already equipped to output the display to the braille without any clunky middle-man software.

It's not an easy choice to make.

I've played your accessible mario. 
I totally agree that there needs to be an option to zoom in to focus on only a section of the image.  Zooming out to get a better feel for the environment around your player is very helpful, but most of the time people will probably only care about what is actually close to you.  How is the braille output handled that prevents it from being done in BGT?  Is it like using the screen readers?  I've never used one so I'm totally clueless.