Text/numbers from Nature scenes?

[Tim Nettleton]

I found a site that uses tesseract and it does VERY well with nature and numbers.

When I use tesseract I do NOT get the same results that they do.

 

I’ve attached an example image that clearly I need to get 1433 for this team member.

When I use your OCR(https://www.imagetotext.info/) it says “AMERICAN FAMILY INSURANC 1433” which is great!

 

When I run tesseract, I get trash on the same image:

 

c:\>tesseract.exe 12749691.jpg stdout -l eng --psm 6 --oem 3

we As eee ┬╗

Ate ae

é FAS

; Z cae f .

if\ iy

i * ΓÇÖ .

| TPE

xX * dp

 

What are we doing wrong? 

Do I need to run a program before tesseract to isolate areas?

What are we missing?

Any help would be great!

 

Thanks,

 

Tim Net

[Rolando José Torres Sánchez]

BUeno creo que serviria mucho nos compartieras la imagen que presenta los problemas, y la version de tesseract que haz usado.de esa forma creo que podemos repetir el error y quizas podemos ayudarte mas. Yo creo que puede deberse a la segmentacion, pues mi idea es que Tesseract, usa diccionario y ancho de los caracteres para calcular estadisticamente la palabra mas exacta en cuanto a los anchos de las letras.

Como los numeros no tienen tales diccionarios, tengo mi teoria que tesseract falla mas cuando trata de leer simbolos y numeros que no estan en ningun tesauro ni diccionario.

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/41f21b4d-e676-4ca9-8ae3-f2d7fd5d09dan%40googlegroups.com.

[Zdenko Podobny]

This is English speaking forum. Please send you message in English.

Zdenko

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAPPP_Bf%2BLTrNXSBRBmDudiOZmrrDvH6Y5wf-qhfGQMgXbPApRA%40mail.gmail.com.