Actually, the parameters would be incredibly difficult. Because there is a good deal of creativity and game insight that goes into inventing a map. It is likely that the program would generate very similar and often stupid maps to play on. The texturing and deals would also be problematic. Map makers will generally range test things and design cliffs and terrain exactly so. Because they first have in mind the kinds of tactics which will be used on their maps.
The height map generator would need to reduce the terrain style down to a fractal algorithum, which would be able to reproduce "real" looking terrain (that is the hard part). One sollution would be to create a highly detailed 81x81 map for each titleset. Then generate a 20x20 map by cropping this map down to the requisite size. Or alternatively, you would need a height map library. Texturing and decals would need to be stytlised, with fractal mathematics.
Starting marker locations would need to be generated, and mex/hydro markers using a series of relative distance templates. With the same template being adopted for each player. The map could also generate a fair distribution of expansion mex sites. It is very likely that all the maps generated would need perfect symmetry for exact balance. However, these advantages would be mitigated by both players having no map knowledge... The generator would need to ensure that mex sites are not placed on unbuildable terrain, and that pathing and access are equivalent. Because these random errors would really spoil a match.
In short. Creating a random map generator is about 100 times more difficult then just making the maps in the first place.
***
However, I have often wondered what a tornament might be like if it featured entierly new maps. With the participants having no contact with the map makers, and launching the game without the ability to study the map beforehand.