Skanerlangan PDF Pipeline

PDF keldi
   |
   v
[Text layer bormi?] ---- HA ---> pdfplumber bilan matnni olamiz
   |                             (bepul, tez, 100% aniq)
   YO'Q
   |
   v
[Har sahifani rasmga aylantirish]  pdf2image, 300 DPI
   |
   v
[Preprocessing]                    3-bobdagi funksiya
   |
   v
[OCR]                              Tesseract (tez) 
   |
   v
[Confidence tekshiruv] -- past --> VLM bilan qayta o'qish
   |
   v
[Tozalash] -> chunking -> embedding (RAG pipeline davom etadi)

import pdfplumber

def is_scanned(pdf_path, min_chars=100):
    """PDF da matn qatlami bormi? Birinchi 3 sahifa yetarli."""
    with pdfplumber.open(pdf_path) as pdf:
        for page in pdf.pages[:3]:
            text = page.extract_text() or ""
            if len(text.strip()) >= min_chars:
                return False       # matn bor — digital PDF
    return True                    # 3 sahifada ham matn yo'q — skanerlangan

sudo apt install poppler-utils    # pdf2image ning motori
pip install pdf2image

from pdf2image import convert_from_path

# Har sahifa PIL Image bo'lib qaytadi
pages = convert_from_path("scanned.pdf", dpi=300)
print(f"{len(pages)} sahifa")

from pdf2image import pdfinfo_from_path, convert_from_path

info = pdfinfo_from_path(pdf_path)
total = info["Pages"]

for start in range(1, total + 1, 10):          # 10 sahifadan
    pages = convert_from_path(pdf_path, dpi=300,
                              first_page=start,
                              last_page=min(start + 9, total))
    for page_img in pages:
        process(page_img)                       # OCR + saqlash

import pdfplumber
import pytesseract
from pdf2image import convert_from_path

def extract_pdf_text(pdf_path):
    """PDF -> matn. Digital bo'lsa to'g'ridan-to'g'ri, bo'lmasa OCR."""
    if not is_scanned(pdf_path):
        with pdfplumber.open(pdf_path) as pdf:
            return "\n\n".join(p.extract_text() or "" for p in pdf.pages)

    pages = convert_from_path(pdf_path, dpi=300)
    texts = []
    for i, page_img in enumerate(pages, 1):
        text = pytesseract.image_to_string(page_img, lang="rus+uzb")
        texts.append(text)
        print(f"Sahifa {i}/{len(pages)} OCR qilindi")
    return "\n\n".join(texts)

Hamma sahifa    ->  Tesseract     (~1 s, bepul, confidence beradi)
                        |
        conf >= 60      |      conf < 60
        ishlatamiz      |      Qwen2.5-VL ga yuboramiz
                        |      (~10 s, kuchli, qiyin sahifalar uchun)