It's not faster to code using one letter variable names, for obvious reasons. Because you're bounded by time to understand and make decisions, and not by how long it takes to press a few keys.
I don't advocate teaching more slowly, but teaching more effectively. Teaching "top down" is more effective, and not widely used, from what I've seen.
In your example you describe a multi-year period where basic examples are given in early childhood, and then made more generic over time. That's similar to what I'm describing, but not really the same.
I'm describing the same methodology but done repeatedly with any new domain being learned.
When you learn a new concept in computer vision, this methodology can still apply. Sure, you can throw the formula to do some image transformation at somebody and have them go at it. But showing it being used, explaining the logic behind it in natural language and so forth will convey the idea much more effectively.
The "bad" math education that turns so many people off tends to be more of a "bottom up" approach, solve 10 generic and meaningless problems using this formula.