Problem: Map a large volume of something that has distinct color.

Solution » High Level: * Highlight a sample square * Send a wanderer around in a perimeter * When the barrier is crossed, stop * Have the end user review

Solution » Low Level [LONG]: public main { //Describe sample square Let the user draw a square //Use sample square int minR, maxR, minB, maxB, minG, maxG; set all the mins to max value and all the max values to 0

for(every pixel in the square)
{
	does it need to be change one of the above 6 ints?
}

//Explore
while(endCondition not triggered)
{
	walk around the perimieter and send it to addSquare()
	if the addSquare says add it, send it to approvedSpots[][]
	else
	{
		ignore it
		but in future versions subdivide the square into 4 areas recursivley to minSizeOfArea;
	}
}

//Encase the area
Find our approved area and go to the top most point from the center.
Drop a point down
Go through the area, if 2 points are on the same line and not necessary, delete it. }

bool addSquare(all 6 rgb values, satteliteArea[][]) { //Filter out the blatantly false average out satteliteAreaRGB if satteliteArea violates the 6 rgb areas, return false.

//Check the square
for all the pixels, how many don't belong
percentDontBelong=numberDontBelong/howmany are in the satteliteArea
    Send it }

Concerns you might have: 1. Automation in OSM? The human has to: Define the sample and approve the result. 2. Junk Data: This would be a problem with a userbase, not a computational problem. We merely create tools. 3. What about non-90 degree angles? I agree that this would be a problem, I would hope someone made a tool that acts like the blender sculpt brush of smooth and bump, but for OSM.

Production: 1. Someone might have made this sort of thing already and I just don’t know about it? If not in JOSM, than in something that costs money. 2. I know that describing a solution is easy and that I did not do the hard part such as the following. How to draw a sample area and convert the pixels into an array, than how to convert the result back.

Edit: My solution (that someone has probably already though of) for non-grid areas is the pin and line model.

Solution » High Level We walk out to the North-most pixel that is compliant with our initial model. We use our string to go around in a circle as far as we can. If alot of errors occur, we place a new pin down where we left off. Repeat the process until we are back to our start.