Detecting Blurs or Sharp Areas

My friend (DaliGiger on FilterForge) has been interested in making a "Focus stacking" filter. It takes multiple photographs with narrow focus ranges and detects areas that are the sharpest. It then attempts to stitch the sharpest area, thereby creating a fully focused image. With FilterForge 3 we have the ability to input multiple images, which is where this whole challenge came out of.

So here's a community question. Are there super efficient ways to find blurry areas and sharp areas using FilterForge? I attempted this challenge myself, I thought I'd throw the method which I came up with here. And I am very sure there are better ways to do this.

The current method I tried uses a very simplistic approach. The image goes through a simple edge detection where the logic goes: if the image is sharp, then there will be loads of edges. If the image is blurry, there will be less edges. Hence: sharp areas will be denser than blurry areas.

So the blur component that comes after the edge detection is specifically for finding that "density." The darker areas mean that the area is probably sharp. The lighter areas would mean that the area is probably blurry. And the rest is just creating a mask with feathering on it for compositing two images.

Though this method doesn't work 100%, and its slow. And I thought maybe the forum has a better solution for this issue. Just wanted to hear thoughts.

This is a problem I've been trying to solve several times.

The main problem is that you can't really use a fixed radius approach. You need access to variable frequency information (a blur only gives you fixed information on a specific frequency). A variable solution requires a transform into frequency domain, which is not currently possible to create in FF (maybe if we get bitmap based scripting it can be done).

But it could probably be faked to suffice with high passes - do you have a sample focus image set?