TextRecognizer reads character `<` as either letters S, C, E or K

[rjmangubat23]

Hello,

TextRecognizer reads MRZ character < either letters S, C, E or K.

Sample below:

Our app is a MRZ scanner which means this issue gets us wrong results especially for givenNames which do not have a check digit (helps validates the data) from the MRZ string scanned.

Currently we have used and tested for both stable repos but it still happens

• TextRecognition - com.google.android.gms:play-services-mlkit-text-recognition:17.0.0

• TextRecognition v2 - com.google.mlkit:text-recognition:16.0.0-beta1

  May I ask if someone else has encountered same problem? or how we should properly approach this issue we are having?

Thank you in advance! :D

[zongmins]

Based on the video and offline testing with Overlay, OCR is fairly accurate on the text below the image but not the text in the image. In addition to the confusion on [<SCEK], there are recognizer errors elsewhere in the passport image. The recognition errors are most likely because the recognizer model wasn't trained with the passport font (monospace font OCR-B).

In the interim, users might try using a higher resolution image, or running recognition multiple times on a video feed to identify which results are stable.

[rjmangubat23]

@zongmins @bcdj thank you for replying. I followed your advice and it did improve it however there are still times that it does happen. May I know where I can ask for improved support for passport font? or is there a way in modifying the recognizer model?

Thank you again!

[bcdj]

You are at the right place for requesting improved support for passport font. :-) We have forwarded your request to the model training team and will continue to work on improving our model. No ETA can be provided at this moment.

[rjmangubat23]

@bcdj may I ask if there are any updates on improved support for passport font in the latest version?

Thank you in advance!