使用 PyMuPDF python 从 PDF 中有效提取突出显示的部分？

Question

我有一个用例，我必须从 PDF 文档中突出显示表格，然后使用 python 提取突出显示的部分。一旦突出显示，我必须将提取的部分转换为数据框，使数据框看起来像这样：

name      |   value
apple inc     0.84
google inc    0.95

我已经实现了从 PDF 中提取突出显示的文本的逻辑。我能够将其提取出来，但是突出显示的部分在我的列表中出现了两次，如果我在每个空格后添加一个逗号，我倾向于省略公司名称，例如：- Apple Inc, Google Inc 是 Apple, Inc ,Google,Inc 我不想要。 这是相同的代码：

from typing import List, Tuple

import fitz  # install with 'pip install pymupdf'
import pandas as pd

def _parse_highlight(annot: fitz.Annot, wordlist: List[Tuple[float, float, float, float, str, int, int, int]]) -> str:
    points = annot.vertices
    quad_count = int(len(points) / 4)
    sentences = []
    for i in range(quad_count):
        # where the highlighted part is
        r = fitz.Quad(points[i * 4 : i * 4 + 4]).rect

        words = [w for w in wordlist if fitz.Rect(w[:4]).intersects(r)]
        sentences.append(" ".join(w[4] for w in words))
    sentence = " ".join(sentences)
    return sentence


def handle_page(page):
    wordlist = page.getText("words")  # list of words on page
    wordlist.sort(key=lambda w: (w[3], w[0]))  # ascending y, then x

    highlights = []
    annot = page.firstAnnot
    while annot:
        if annot.type[0] == 8:
            highlights.append(_parse_highlight(annot, wordlist))
        annot = annot.next
    return highlights


def main(filepath: str) -> List:
    doc = fitz.open(filepath)

    highlights = []
    for page in doc:
        highlights += handle_page(page)

    #print(highlights)
    highlighted_text = [",".join(i.split(" ")) for i in highlights] # adds a comma in place of spaces
    print(highlighted_text)
    # df = pd.DataFrame(highlights)
    # print(df.head())


if __name__ == "__main__":
    (main("pdfname.pdf"))

我得到的输出是：

['Security,Name,%,to,Net,Assets*,Assets*,DEBENtURES,0.04,Britannia,Industries,Ltd.,0.04,Britannia,Industries,Ltd.,0.04,EQUity,&,EQUity,RELAtED,96.83,EQUity,&,EQUity,RELAtED,96.83,HDFC,Ban
k,Ltd.,6.98,HDFC,Bank,Ltd.,6.98,ICICI,Bank,Ltd.,4.82,ICICI,Bank,Ltd.,4.82,Infosys,Ltd.,4.37,Infosys,Ltd.,4.37,Reliance,Industries,Ltd.,4.05,Reliance,Industries,Ltd.,4.05,Bajaj\tFinance\tL
td.,3.82,Bajaj\tFinance\tLtd.,3.82,Housing,Development,Finance,Corpn.,Ltd.,3.23,Housing,Development,Finance,Corpn.,Ltd.,3.23,Grindwell,Norton,Ltd.,3.22,SRF,Ltd.,3.22,SRF,Ltd.,3.22,Sun,Pha
rmaceutical,Industries,Ltd.,2.85,Sun,Pharmaceutical,Industries,Ltd.,2.85,Bharti,Airtel,Ltd.,2.82,DLF,Ltd.,2.64,DLF,Ltd.,2.64,Ultratech,Cement,Ltd.,2.62,Ultratech,Cement,Ltd.,2.62,SKF,Indi
a,Ltd.,2.45,Crompton,Greaves,Consumer,Electricals,Ltd.,2.42,Crompton,Greaves,Consumer,Electricals,Ltd.,2.42,Avenue,Supermarts,Ltd.,2.41,Avenue,Supermarts,Ltd.,2.41,Axis,Bank,Ltd.,2.41,ABB
,India,Ltd.,2.35,ABB,India,Ltd.,2.35,Titan,Co.,Ltd.,2.29,Titan,Co.,Ltd.,2.29,Kotak,Mahindra,Bank,Ltd.,2.09,Cipla,Ltd.,2.05,Cipla,Ltd.,2.05,Laurus,Labs,Ltd.,2.04,Laurus,Labs,Ltd.,2.04,Wipr
o,Ltd.,1.77,Happiest,Minds,Technologies,Ltd.,1.68,Happiest,Minds,Technologies,Ltd.,1.68,Canara,Bank,1.67,Canara,Bank,1.67,Shree,Cement,Ltd.,1.63,Security,Name,%,to,Net,Assets*,Assets*,Mah
indra,&,Mahindra,Ltd.,1.59,Pidilite,Industries,Ltd.,1.50,Pidilite,Industries,Ltd.,1.50,ICICI,Lombard,General,Insurance,Co.,Ltd.,1.48,ICICI,Lombard,General,Insurance,Co.,Ltd.,1.48,Cholaman
dalam,Investment,&,Finance,Co.,Ltd.,1.45,Cholamandalam,Investment,&,Finance,Co.,Ltd.,1.45,Tech,Mahindra,Ltd.,1.35,Tech,Mahindra,Ltd.,1.35,State,Bank,of,India,1.31,State,Bank,of,India,1.31
,Hindustan,Unilever,Ltd.,1.30,Hindustan,Unilever,Ltd.,1.30,Vardhman,Textiles,Ltd.,1.30,Vardhman,Textiles,Ltd.,1.30,Larsen,&,Toubro,Ltd.,1.24,Larsen,&,Toubro,Ltd.,1.24,Dabur,India,Ltd.,1.2
2,Neogen,Chemicals,Ltd.,1.10,Neogen,Chemicals,Ltd.,1.10,Eicher,Motors,Ltd.,1.09,Eicher,Motors,Ltd.,1.09,Thermax,Ltd.,1.08,TATA,Consultancy,Services,Ltd.,1.05,TATA,Consultancy,Services,Ltd
.,1.05,Indian,Railway,Catering,&,Tourism,Corpn.,Ltd.,0.98,Indian,Railway,Catering,&,Tourism,Corpn.,Ltd.,0.98,Firstsource,Solutions,Ltd.,0.97,Nestle,India,Ltd.,0.86,Nestle,India,Ltd.,0.86,
Asian,Paints,Ltd.,0.84,Asian,Paints,Ltd.,0.84,Welspun,India,Ltd.,0.72,IndusInd,Bank,Ltd.,0.63,IndusInd,Bank,Ltd.,0.63,SBI,Life,Insurance,Co.,Ltd.,0.50,SBI,Life,Insurance,Co.,Ltd.,0.50,Dee
pak,Nitrite,Ltd.,0.46,Adani,Ports,and,Special,Economic,Zone,Ltd.,0.36,Adani,Ports,and,Special,Economic,Zone,Ltd.,0.36,Gateway,Distriparks,Ltd.,0.33,Gateway,Distriparks,Ltd.,0.33,Bharat,Fo
rge,Ltd.,0.22,tREPS,on,G-Sec,or,t-Bills,2.81,tREPS,on,G-Sec,or,t-Bills,2.81,Cash,&,Cash,Receivables,0.32,Cash,&,Cash,Receivables,0.32,tOtAL']

总结一下：如何确保在从 pdf 中提取突出显示的文本时不会出现重复项和 2. 如何确保值与在 pdf 中看到的值相同，例如 Google Inc、Apple Inc不像 Google,Inc,Apple,Inc.

请帮帮我。

编辑：- 从集合库中实现了计数器，仍然无法删除多余的名称。

from collections import Counter
def handle_page(page):
    wordlist = page.getText("words")  # list of words on page
    new_wordlist = sorted(Counter(wordlist), key=lambda w: (w[3], w[0]))
    #wordlist.sort(key=lambda w: (w[3], w[0]))  # ascending y, then x

    highlights = []
    annot = page.firstAnnot
    while annot:
        if annot.type[0] == 8:
            highlights.append(_parse_highlight(annot, new_wordlist))
        annot = annot.next
    return highlights

EDIT-2：实现了这个逻辑，现在我可以从列表中删除多余的单词，但是像（“LTD”）这样的单词正在被删除。

def _parse_highlight(annot: fitz.Annot, wordlist: List[Tuple[float, float, float, float, str, int, int, int]]) -> str:
    points = annot.vertices
    quad_count = int(len(points) / 4)
    sentences = []
    for i in range(quad_count):
        # where the highlighted part is
        r = fitz.Quad(points[i * 4 : i * 4 + 4]).rect

        words = [w for w in wordlist if fitz.Rect(w[:4]).intersects(r)]
        sentences.append(" ".join(w[4] for w in words))
    sentence = " ".join(sentences)
    string_split = sentence.split()
    sent = " ".join(sorted(set(string_split), key=string_split.index))
    return sent

Answer 1

使用 re.sub() 解决了这个问题

def _parse_highlight(annot: fitz.Annot, wordlist: List[Tuple[float, float, float, float, str, int, int, int]]) -> str:
    points = annot.vertices
    quad_count = int(len(points) / 4)
    sentences = []
    for i in range(quad_count):
        # where the highlighted part is
        r = fitz.Quad(points[i * 4 : i * 4 + 4]).rect

        words = [w for w in wordlist if fitz.Rect(w[:4]).intersects(r)]
        sentences.append(" ".join(w[4] for w in words))
    sentence = " ".join(sentences)
    string_split = sentence.split(",")
    sent = " ".join(sorted(set(string_split), key=string_split.index))
    output = re.sub(r'\b(\d+(?:\.\d+)?)\b', r'\1,', sent)
    return output