Correct license plate with wrong letter read

[Lucas Doná Sfalcin]

Good afternoon everyone! I'm trying to reduce the amount of garbage (incorrect plates) in my database by doing the following:
I look at the size of the plate with "warpedBox". Then I compare the two plates. The larger plate is theoretically the closest photo, so it probably has a better chance of being the correct one, but sometimes it isn't, like in the two attached photos.
Is there a better way to know which one would be the correct plate between these two photos?

Why does this happen?
I will have a database receiving license plate recognition from a thousand vehicles.

Thank you.

[Mamadou DIOP]

Hi,

When I try with https://www.doubango.org/webapps/alpr/ both images are correctly recognised. I guess you’re using mobile version (quantised models) which are less accurate.

Licenses plates from Brazil (https://en.wikipedia.org/wiki/Vehicle_registration_plates_of_Brazil) have a fixed format depending on the version. 

New plates (https://en.wikipedia.org/wiki/Vehicle_registration_plates_of_Brazil#/media/File:Placas_Mercosul_2022.jpg) have format LLLNLNN

Old plates (https://en.wikipedia.org/wiki/Vehicle_registration_plates_of_Brazil#/media/File:Placas_de_ve%C3%ADculos_no_Brasil_2009.jpg) have format LLLNNNN

1/ Use following regex: ([A-Z]{3}\d{1}[A-Z]{1}\d{2})|([A-Z]{3}\d{4})

2/ You can also use score

--
You received this message because you are subscribed to the Google Groups "doubango-ai" group.
To unsubscribe from this group and stop receiving emails from it, send an email to doubango-ai...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/doubango-ai/63eac56b-4459-4a5d-a0a2-eb1d02966cadn%40googlegroups.com.
<correct_PPB8268.jpg><error_PPB8266.jpg>

[Lucas Doná Sfalcin]

I use Alpr Doubango with a Raspberry pi 3, Raspberry pi Os Bookworm and Python with openCV.
In this specific case, the 2 photos attached, the score of the wrong plate was higher, so my algorithm would choose the wrong photo as well.

[Mamadou DIOP]

use regex first then score

share the result and score you get for each image

To view this discussion visit https://groups.google.com/d/msgid/doubango-ai/4c7b4aee-a854-4e61-b510-2d567751f9f4n%40googlegroups.com.