alkasm / ifm-challenge Goto Github PK

My first idea for a confidence value was to cross-check the images with each other by finding the correlation of all detected labels, but this requires multiple detections to have taken place. So instead, I'm going to opt for creating an average label image (just scaling the labels to the same size and using a mean filter across the images) and using that as a template for correlation with the found labels. If this image is defined with good quality labels, then scaling them to the size of the detected label should give a decent match with good quality labels, and not as good with low quality resolution labels, and should give a bad result with no label detection.