What kind of image preprocessing may improve Vision accuracy?

Hello. We are doing barcodes and article IDs recognition using Vision - specifically VNDetectBarcodesRequest and VNRecognizeTextRequest. Previously we were doing our own preprocessing of images like cropping the region of interest and applying monochrome filter. However, I noticed that it actually reduces the accuracy in some cases and in WWDC session it was said that Vision also does some preprocessing. Should we do preprocessing at all in this case? What kind of custom image preprocessing might be useful for this task? Are there any tips how to increase the accuracy of Barcode/Text recognition?