Reg. google cloud vision api

I am trying to read the hand-written data on hot rolled coils using python and google-cloud-vision api. But what i have observed that when the data is written in slightly inclined manner the vision api completely fails. For eg if the data is written as it completely fails and gives some absurd output even though visibly the data should be ok to read.

However if the data is written as the api reads it correctly.
Should any pre-processing for the image needs to be done so that api reads it correctly?

Image pre-processing is a good practice on the client side to improve efficiency, accuracy and response time. In fact, it would be particularly important to remove as much noise as possible.The Vision API, however, uses machine learning algorithms that are continuously improved. As far as my knowledge about image pre-processing techniques, this task is more easily accomplished manually than using standard or advanced image pre-processing. On this Google cloud vision page, you can find a good set of recommendations.