Yeah, the subject line pretty much asks what I'm trying to ask here...
I've heard of the technique but don't know how they do it.
Btw, is this one of the main gaming dev forums out there? I've looked at a few others and this one seems more polished I would say. I'm a student and am looking for good discussion places. Thanks for the help. This site looks really decent.
You can create a texture for depth from an image, but it doesn't always work right. It just uses the dark and light places on the texture to map depth. There are software programs to do it. Gimp has a plugin for it. Usually for normal maps, though, a depth texture is baked from a light map. Programs like Blender do it, or there are specific programs which also do it which I can't remember the name. You can also do maps for alpha and specularity. I think that's what you are asking about. They're done in modeling programs but not from one image. They are combined in the engine.
I think it's basically guesswork / eyeballing. Oftentimes textures that are from a photo reference will have some lighting baked into them, so you can kinda get some kind of approximation of the height map by looking at contrast, like fireside says. But it's difficult to tease that out from actual changes in the material color. For instance, a sidewalk with cracks between the concrete tiles will have dirt accumulate in the cracks so they'll both be darker in color, and darker because less light gets down in there.
For getting actual height/normal map, you need to take a photo of the same surface (from same viewpoint) with different conditions of lighting
- if you have all this, you can actually build correct normal map and thus correct height map (as long as you know the depth difference between most distant point and less distant point in viewport (actually this one is possible to guess without any actual measurement).
In real development, there is no time or money to do the previous one. So you actually hand-create the height maps (or try to "fake" extract it somehow) and normals map (from height map).
So you actually hand-create the height maps (or try to "fake" extract it somehow) and normals map (from height map).
Alternatively, you actually model the details and bake them to texture/normal/specularity-maps.
Or buy a 3D camera, they are not that expensive now.