Images in output as slices of original scan
I am not sure about this, but I have the impression that the images output of free text question are slices not of the original image but of some intermediate processing format. I noticed this because some barcodes in the output images cannot be read, also if in the original scan they can. Probably for written text this is not even noticeable however seems barcode cannot live with it... I have a success result of 244/659 = 37% in my tests for reading barcodes.