When looking upon a wooded hill, we can choose to see the "forest or the trees."  Nevertheless, we generally see one or the other on default, following a combination of various heuristic rules (apparent size relative to visual field; contrast of size, color, and shape; sharpness; etc.)  Simply put, we will likely focus on the level of detail that stands out the most.

A black ninja outfit might be excellent camouflage at night under a new moon, but it will stick out like a sore thumb on an open street in the day time.

However talented at stealth an wood elf might be, they are probably going to stick out like a sore thumb in the streets of a dwarven city.

Whether or not something is well camouflaged, or on the other side of things, highly noticeable, often has everything to do with visual environment, which can change dramatically.

The very act of writing a description requires modelling the heuristic rules involved in noticing things.  But how much of the act of noticing things have you automated?  In your automated system, which heuristic rules did you focus on simulating?

I'm currently trying to conceive of a feasible system in which selective attention is based mostly on the size, position, and coded apparant material of objects.  While a green jacket might coded as being made out out wool, and this would be noticeable by looking directly at the object, it's coded apparent material might be something like "vegetation".  Thus, when in a room dominated by stone objects, the jacket would be very noticeable, but in a room dominated by vegetation the jacket would be much less noticeable.

The materials that dominate a room could be determined percentage-wise based on preset value (determined by the overall environment, such as a forest) and then modified by the sizes of the objects of each material in the room.  If I use a coordinate-system instead of "rooms," I will have to include position into the equation; but I'm not ready to think that through yet.

I'm not sure this is a feasible way to go. After all, what works with the 'green wool jacket', isn't very likely to be true with the 'blue wool hat'. Even though fabricated out of the same materials, it is highly likely that you'd see it from a mile away in any kind of terrain.

My suggestion is to focus on object size, lighting (infravision could have an effect), perception of the player (and a 'search command' for looking closer in a spot). You could try and use a 'brightness' factor in the objects as well, so a shiny ring or dagger would be easier to notice, than for instance a small leather sling. A room could have it's own environmental setting, based on the materials present, the time of day, the number of lightsources in the vicinity, and depending on the difference between the object brightness and the environment, the object could be more or less noticable.

Anyway - just my two cents.

Welcor