Training/Altering the bounding box

Hi dragos@,

Thank you for choosing AWS Textract and sorry to hear that you are facing issues with the detection quality. It would be helpful (only if it's possible) if you can share a sample document with us to provide better guidance on what would be the most appropriate API to use between AWS Textract and AWS Rekognition.

We do support PDFs in asynchronous APIs for AWS Textract and black and white images (in JPEG/PNG formats) in both synchronous and asynchronous APIs.

Regards,
Anand

Hi dhandhaniaataws@,

Thank you for reaching out to me regarding this problem.
https://www.google.com/search?q=buletin&rlz=1C1CHBF_enRO883RO883&sxsrf=ALeKk018EZUkVvdgdUudcLfYmfIWhj5Uow:1593414526596&source=lnms&tbm=isch&sa=X&ved=2ahUKEwjgrdnWu6bqAhWLp4sKHchwBkYQ_AUoAXoECAsQAw&biw=1536&bih=722#imgrc=BSbMqw4Pcmm06M
This is a google image with an identification card that I use for my project. The problem lies in the last line, the one starting with "KT...". This line is broken into Words but some Words use the same characters. For example in the google image the last line could be identified as : "KT2100_2100_" or the last character is not identified.
And regarding the input format, I use a pdf with a scanned photo of the identification card in black and white.
I hope this information helps you.

Thank you for your time!

Hi dragos@,

Thanks for the reply, I run the image you provided through our system. However, I cannot reproduce the issue you are facing and our system seems to produce the right result (KT210012<) with the last line that starts with KT.

Could you share more detail information and sample documents if possible so that we can further assist you?

Thanks.