How to extract text from image #32

Activity

1. JorjMcKie commented on Jan 21, 2022

   

   JorjMcKie

   on Jan 21, 2022

   Collaborator

   More actions

   First problem is, that it is no image.

   >>> import fitz
   >>> doc=fitz.open("BOURGEOISFacture.21053886.pdf")
   >>> page=doc[0]
   >>> from pprint import pprint
   >>> # no standard images there:
   >>> pprint(page.get_images())
   []
   >>> # neither embedded images:
   >>> pprint(page.get_image_info())
   []

   And you will have determined that what seems to be text neither is text.

   >>> page.get_text(sort=True)  # "sort" ensures text visible at the top indeed comes first
   BENOIT CASTEL MENILMONTANT 
   Boulangerie Pâtisserie 
   150 rue de Ménilmontant 
   75020 PARIS 
   06 21 08 29 68
   Tél. Client :
   FACTURE
   /
   1
   ... (more data) ...

   So the apparent text must be encoded as drawing primitives, like a capital "B" drawn as a vertical line | followed by two little semi-circles above each other, etc.
   The only way is therefore OCRing this area. Let's find a suitable sub-rectangle: left of "BENOIT CASTEL MENILMONTANT" and above the first "FACTURE":

   >>> rl1 = page.search_for("BENOIT CASTEL MENILMONTANT")
   >>> len(rl1)
   3
   >>> rl2 = page.search_for("FACTURE")
   >>> len(rl2)
   2
   >>> # sort both rectangle lists to be sure: vertical, then horizontal
   >>> rl1.sort(key=lambda r: (r.y1, r.x0))
   >>> rl2.sort(key=lambda r: (r.y1, r.x0))
   >>> # right border is left or first rl1 rect:
   >>> rborder = rl1[0].x0
   >>> # bottom is top coord of first rect of rl2:
   >>> bottom = rl2[0].y0
   >>> # define sub rect to OCR:
   >>> clip = fitz.Rect(0,0,rborder,bottom)
   >>> clip
   Rect(0.0, 0.0, 339.3900146484375, 220.2386474609375)
   >>> # make a pixmap of that rect:
   >>> pix = page.get_pixmap(dpi=300,clip=clip)
   >>> # make a new 1 page PDF with OCRed text
   >>> pdfbytes = pix.pdfocr_tobytes()
   >>> ocrpdf = fitz.open("pdf", pdfbytes)
   >>> ocrpage = ocrpdf[0]
   >>> print(ocrpage.get_text())
   Bourgeois Fréres S.A.S au Capital de 330 000 Euro
   77510 VERDELOT
   TEL
   : 01 64 04 81 04
   FAX
   : 01 64 04 81 43
   SITE INTERNET
   : WWW.MOULINS-BOURGEOIS.COM
   N° SIRET 746 050 087 00012
   R.C. MEAUXB 746 050 087
   CODE APE: 1061A
   N° de TVA INTRACOMMUNAUTAIRE
   : FR15746050087
   
   >>> 
2. JorjMcKie commented on Jan 21, 2022

   

   JorjMcKie

   on Jan 21, 2022

   Collaborator

   More actions

   Of course that OCRed text could also have been extracted with coordinates, e.g. using ocrpage.get_text("dict").
   Those coordinates obviously are relative to that ocrpage's dimensions. If required translate them back to the original page's positions using the matrix mat = ocrpage.rect.torect(clip).
3. Larbo53 commented on Jan 21, 2022

   

   Larbo53

   on Jan 21, 2022 via email

   Author

   More actions

   Thanks a lot Which package must be imported to works? Sincerely Yves Larbodiere EVARD 3, rue des Courtes Terres 95220 HERBLAY portable : 07 81 08 41 00 mail : ***@***.***

   …

   Le 21 janv. 2022 à 16:24, Jorj X. McKie ***@***.***> a écrit : Of course that OCRed text could also have been extracted with coordinates, e.g. using ocrpage.get_text("dict"). Those coordinates obviously are relative to that ocrpage's dimensions. If required translate them back to the original page's positions using the matrix mat = ocrpage.rect.torect(clip). — Reply to this email directly, view it on GitHub <#32 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AU2RHDFZP4VARUR5INAOVT3UXF3BBANCNFSM5MPUEXAQ>. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.
4. pymupdf

   locked and limited conversation to collaborators on Jan 21, 2022
5. JorjMcKie

   converted this issue into a discussion #34 on Jan 21, 2022