So the way which correspondence problem is made most obvious is in terms of Random Dot Stereograms. For many people that's really the definition of a correspondence problem. Let me tell you a little bit about that. A person who really developed a Random Dot Stereograms in the 1950s, was Bela Julsez, who's not the first to actually discover Random Dots Stereograms but was the first to put it in digital form allowing many, many researchers to make Random Dot Stereograms with computer algorithms. And test them in a whole variety of psycho-physical circumstances which Julesz did himself and many others followed and many observation that rove. So what is a Random Dot Stereograms and how is it pertain to the correspondence problem? Well, first of all let's describe the phenomenon itself and I think those of you who can do what is called free fusion, who have the ability to look through the computer screen and you all can do this with proctor so I promise you I didn't think I could it until I got interested in this and practiced a little bit. If you fuse these two images, in free fusion, just by looking through the computer screen, you will see an object that pops out in three dimensions that looks like it is behind the surface that the rest of the dots comprise. And if you have trouble doing that, I suggest you take a piece of cardboard. Just an envelope or something that blocks the two eye views and put the piece of cardboard between the two stereograms and put your nose in front of that. Then you'll see the right one with one eye, the right eye, and the left one with the left eye. And you'll automatically fuse them and the object will eventually pop out. So it's fun to do that. That's the basis of the amusement that comes from stereograms and autostereograms that were 10 or 15 years ago, very popular in posters, in newspaper presentations and so on. So it's fun to look at these things but it has, in addition to being fun, a deep significance, in terms of explaining stereoscopic vision. So how is a stereogram like this made? Well, let's consider the right eye view. The whole right eye view image is just a pattern of random dots made as you are showed easily with an algorithm that you can now use on any laptop. What the experimenter does or anyone who wants to play with this does, is to pick an area of the scene. Doesn't have to be a square it could be anything. And in the view of the other eye, take that same square, And shift it a little bit. Just a few degrees. So that leaves a gap where it's been shifted. And a cover up area where it includes the set of random dots that were there in the view of the other eye. Well, now, what you have is two sets of random dots in the two images and those sets of random dots are perfectly camouflage unless you use stereoscopic vision, so if you look at these with one eye, you'll never see the object that's there. If you look at them with two eyes you have to fuse the left and the right eye images and then the object will pop out. Well, breaking camouflage and much of the work that was done on this early on before US got into the act and after, was funded by the military. Why? Because breaking camouflage is an important thing. So if you think of these sets of random dots as aerial views of camouflage tanks on the ground or something like that. If you think of using stereoscopy or images that have been created by stereoscopic cameras at an overflowing plane, you can imagine seeing things that you couldn't see otherwise by seeing them in stereoscopic view. Well, I don't think that's of much interest these days but it certainly was in the 1950s. But that's how you make a stereogram, and it's pretty simple, and it can be fun to do. What's the message of the Random Dot Stereogram for the correspondence problem? Well, remember I said before that in disparity whether it's cross disparity or uncrossed disparity that your job is to match p and q [COUGH] in a cross disparity image or match m and n in a [COUGH] view of this that is uncrossed disparity and the question is, how could you do that? That's image matching. You have to figure out someway that you can take the left eye image and the right eye image and match them. And if you think about images, well, maybe it's not so hard to conceive. But if you think about Random Dot Stereograms, it's very hard to conceive because matching this image and this image, which are nothing more than a set of random dots, how could you possibly do that? That's why Random Dot Stereograms have been important in bringing to the forefront the correspondence problem there is a challenge that you can kind of imagine where is around when you're looking at real images. But when you're looking at a completely camouflage object like the shifted square in these two Random Dot Stereograms, there's no way you're going to do that in any simple manner by using image information. Well, there's information in the image all right but it's only stereoscopic information. There's no image about objects and that's really been the salient reason for so much work on Random Dot Stereograms that you get rid of all objects, you pare the problem down to its simpler dimensions. And those simpler dimensions involve the question, how can you match random dots in the left eye view and the right eye view? And that's no easy matter. So as I said, this is one of the problems that simply unresolved in mysterious scout division. It's not really agreed upon to this day how this that the correspondents probably solve, there are many ideas using a complex algorithms that interior phase information, disparity information, other information. But it's just not clear how that's done and or maybe otherwise, I'm thinking about if the people are still unsure how this problem is solved today.
