The most optimal image resolutions are 1280×1024 or 1440×1280 in either landscape or portrait mode.

Much lower resolution can result in inaccurate recognition, and higher resolutions do not improve recognition significantly and will only result in a bigger volume of data.

Image quality (focused and including the full container code) is also equally important. Even though the ConPDS OCR Engine, as part of the image pre-processing phase, is doing Image Noise reduction, Binarization, Image emphasis, and so on, it will make a better starting point for the detection and correct recognition of characters.

Get more details here.