Im making a simple java spring boot app where i upload an image of a cheque on an html page and then the backend uses micr/ocr to read the cheque and only extract the micr line at the bottom. I used tesseract but it gives correct result for some images and incorrect for others. What approach should i take and are there any other free open source sdk to read and identify magnetic ink on cheques i can use?

While using tesseract my approach was crop the bottom half of image and only read the micr line, the problem lies with identifying the characters and distinguishing the numbers correctly.