I wanted to crop out a specific rectangle in a few dozen scanned documents with ImageMagick like this:
Scanning often results in a tiny skew, which would lead to a slightly different rectangle location on every image. It seems that a common preprocessing step when doing OCR is to automatically correct this skew. The Python toolset ocropy for example contains a tool to do this:
convert -crop 1600x1880+100+420 image.nrm.png cropped.png
./ocropus-nlbin image.jpgcreates the file
image.nrm.pngwhich is optimized for OTR and has corrected skew.