Improving the accuracy in number recognition in tesseract-ocr

Question

I recently started using tesseract-ocr with the help of sharp (a node.js image editor). In my algorithm a certain picture is supposed to get resized and cropped by sharp and get