Fixes to how labels are found (and used in cropping) during the binarization step
Previous method was crashing due to labels being rejected by a hard-coded area threshold
New method checks the largest regions in the image, and selects the 3 that best represent a tag based on the ratio of the area of its bounding box (that is defined by its major and minor axes) to the pixels of the tag. also does other stuff like eroding and hole filling to make this easier
Fixes to how labels are found (and used in cropping) during the binarization step