I think I understand the concept been discussed here. The feature wanted I would guess work in the following manor.
1. Acquire requisite photographic source martial and use a tool that would be the landscape terrain modeling version of the Magnetic Lasso tool in Photoshop, which you would acquire 2d profiles from the source photograph.
2. The profiles I would guess would work after a fashion like the profile tool in some CAD software where if you wish to model some part like a cam or other such part you would draw its profile and then lath around an axis to get the desired part: the system as I understand it would go a stage further then this and could be seen as 2d contours describing part of a landscape in 3d space.
2.a. This method would require sophisticated discrimination and inference routines to fill in the missing data (Information not seen be the camera due POV and soon) though it would be largely guess work, and would more than likely be computationally and memory allocation expensive.
3. The 2D profile(s) so gathered would then be sent to the terrain generation engine and the profiles would shape a Power-Fractal or what ever into the desired landscape.
3.a Alternatively the 2d profiles could be gathered outside Terragen in say (Insert name of image editing program here) and then imported into the node designed for purpose and then (3) would apply.
Thats my guess how the requested feature would work and would require considerable development work to implement. Thats my two Cents any away.
Regards to you.
Cyber-Angel