One possibility is to rasterize (e.g. using the ImageRasterizer) the polygons and using the TesseractCaller to do OCR on the result. When rasterizeing, you may want to make sure the polygons are filled, not just outlined.

Hi David !

I've got a rejection message :

Teeseract didn't find any text on this raster

Is there a problem with my parameters ? see attached images

I've never used the TesseractCaller myself, but perhaps you can find some pointers either in the documentation, or in this article: https://www.safe.com/blog/2016/10/ocr-for-fme-now-i-know-my-abc/

Ok.

Thank you very much ! I think the TesseractCaller is the solution to my problem. I'll read the documentation to see about the parameters.

Thanks again and have a nice day !

An alternative solution would be to use the RekognitionConnector with action "Text detection". It will use AWS to perform text detection, and in my experience it is really, really good.

The service might be billed depending on the number of requests, although I find it very reasonable for the quality it offers.