I have a classified satellite image and I want to create a confusion matrix to check its accuracy. The satellite image has homogenous areas (1 landclass label) and complex regions (lots of landclass labels within a region) For the sample points, I want them to be weighted so they're more likely to occur in complex areas where lots of labels occur.
My current plan is to do an edge detection to act as a measure of complexity (areas with more classes should have more edges) and use that to weigh the random sample point, where by an area with a high number of edges is more likely to have a sample point.
How can I do this?
