FORMAT_CONVERSION_MAXCOMPUTE/format_convert/convert_pdf.py

import copy
import io
import os
import re
import sys
from bs4 import BeautifulSoup
sys.path.append(os.path.dirname(__file__) + "/../")
from pdfplumber import PDF
from pdfplumber.table import TableFinder
from pdfplumber.page import Page as pdfPage
from format_convert.convert_tree import _Document, _Page, _Image, _Sentence, _Table, TextBox
import time
from PIL import Image
import traceback
import cv2
import PyPDF2
from PyPDF2 import PdfFileReader, PdfFileWriter
from pdfminer.pdfparser import PDFParser
from pdfminer.pdfdocument import PDFDocument
from pdfminer.pdfpage import PDFPage
from pdfminer.pdfinterp import PDFResourceManager, PDFPageInterpreter
from pdfminer.converter import PDFPageAggregator
from pdfminer.layout import LTTextBoxHorizontal, LAParams, LTFigure, LTImage, LTCurve, LTText, LTChar, LTRect, \
    LTTextBoxVertical, LTLine, LTTextContainer, LTTextLine
from format_convert.utils import judge_error_code, get_platform, LineTable, log, \
    memory_decorator, get_garble_code, get_md5_from_bytes, bytes2np, bbox_iou, get_garble_code2, get_traditional_chinese
import fitz
from format_convert.wrapt_timeout_decorator import timeout
from otr.table_line_pdf import table_line_pdf
# import jieba


@memory_decorator
def pdf2text(path, unique_type_dir):
    return


@timeout(10, timeout_exception=TimeoutError)
def pdf_analyze(interpreter, page, device, page_no):
    pdf_time = time.time()
    interpreter.process_page(page)
    layout = device.get_result()
    log("page_no: " + str(page_no) + " pdf_analyze cost: " + str(time.time() - pdf_time))
    return layout


@timeout(25, timeout_exception=TimeoutError)
def read_pdfminer(path, laparams):
    fp = open(path, 'rb')
    parser = PDFParser(fp)
    doc_pdfminer = PDFDocument(parser)
    rsrcmgr = PDFResourceManager()
    device = PDFPageAggregator(rsrcmgr, laparams=laparams)
    interpreter = PDFPageInterpreter(rsrcmgr, device)
    return doc_pdfminer, device, interpreter


@timeout(15, timeout_exception=TimeoutError)
def read_pymupdf(path):
    return fitz.open(path)


@timeout(15, timeout_exception=TimeoutError)
def read_pypdf2(path):
    doc_pypdf2 = PdfFileReader(path, strict=False)
    doc_pypdf2_new = PdfFileWriter()
    return doc_pypdf2, doc_pypdf2_new


@timeout(25, timeout_exception=TimeoutError, use_signals=False)
def read_pdfplumber(path, laparams):
    fp = open(path, 'rb')
    lt = LineTable()
    doc_top = 0
    doc_pdfplumber = PDF(fp, laparams=laparams.__dict__)
    return lt, doc_top, doc_pdfplumber


class PDFConvert:
    def __init__(self, path, unique_type_dir, need_page_no):
        self._doc = _Document(path)
        self.path = path
        self.unique_type_dir = unique_type_dir
        if not os.path.exists(self.unique_type_dir):
            os.mkdir(self.unique_type_dir)

        # 指定提取的页码范围
        self.need_page_no = need_page_no
        self.start_page_no = None
        self.end_page_no = None
        # 默认使用limit_page_cnt控制，前10页后10页
        if self.need_page_no is None:
            self.limit_page_cnt = 20
        else:
            # 使用start_page_no,end_page_no范围控制，例如2,5
            ss = self.need_page_no.split(',')
            if len(ss) != 2:
                self._doc.error_code = [-14]
            else:
                self.start_page_no = int(ss[0])
                self.end_page_no = int(ss[-1])
                if self.end_page_no == -1:
                    self.end_page_no = 1000000
                self.start_page_no -= 1
                self.end_page_no -= 1
                if self.end_page_no <= self.start_page_no or self.start_page_no < 0 or self.end_page_no < -1:
                    self._doc.error_code = [-14]

        self.packages = ["pdfminer", "PyMuPDF", "PyPDF2", "pdfplumber"]
        self.has_init_pdf = [0] * len(self.packages)

        # 记录图片对象的md5，用于去除大量重复图片
        self.md5_image_obj_list = []

    @memory_decorator
    def init_package(self, package_name):
        # 各个包初始化
        try:
            laparams = LAParams(line_overlap=0.01,
                                char_margin=0.3,
                                line_margin=0.01,
                                word_margin=0.01,
                                boxes_flow=0.1, )
            if package_name == self.packages[0]:
                self.doc_pdfminer, self.device, self.interpreter = read_pdfminer(self.path, laparams)
                self.has_init_pdf[0] = 1

            elif package_name == self.packages[1]:
                self.doc_pymupdf = read_pymupdf(self.path)
                self.has_init_pdf[1] = 1

            elif package_name == self.packages[2]:
                self.doc_pypdf2, self.doc_pypdf2_new = read_pypdf2(self.path)
                self.has_init_pdf[2] = 1

            elif package_name == self.packages[3]:
                self.lt, self.doc_top, self.doc_pdfplumber = read_pdfplumber(self.path, laparams)
                self.has_init_pdf[3] = 1
            else:
                log("Only Support Packages " + str(self.packages))
                raise Exception
        except Exception as e:
            log(package_name + " cannot open pdf!")
            traceback.print_exc()
            self._doc.error_code = [-3]

    def convert(self, limit_page_cnt=20):
        if self.has_init_pdf[0] == 0:
            self.init_package("pdfminer")
        if self._doc.error_code is not None:
            self._doc.error_code = None
            # pdfminer读不了直接转成图片识别
            self.get_all_page_image()
            return

        # 判断是否能读pdf
        try:
            pages = PDFPage.create_pages(self.doc_pdfminer)
            for page in pages:
                break
            pages = list(pages)
        # except pdfminer.psparser.PSEOF as e:
        except:
            # pdfminer 读不了空白页的对象，直接使用pymupdf转换出的图片进行ocr识别
            log("pdfminer read failed! read by pymupdf!")
            traceback.print_exc()
            try:
                self.get_all_page_image()
                return
            except:
                traceback.print_exc()
                log("use pymupdf read failed!")
                self._doc.error_code = [-3]
                return

        # 每一页进行处理
        pages = PDFPage.create_pages(self.doc_pdfminer)
        pages = list(pages)
        page_count = len(pages)
        page_no = 0
        for page in pages:
            # 指定pdf页码
            if self.start_page_no is not None and self.end_page_no is not None:
                if page_count < self.end_page_no:
                    self.end_page_no = page_count
                if page_no < self.start_page_no or page_no >= self.end_page_no:
                    page_no += 1
                    continue
            # 限制pdf页数，只取前后各10页
            else:
                if page_count > limit_page_cnt and int(limit_page_cnt/2) <= page_no < page_count - int(limit_page_cnt/2):
                    page_no += 1
                    continue

            # 解析单页
            start_time = time.time()
            self._page = _Page(page, page_no)
            self.convert_page(page, page_no)
            log('convert_page page_no: ' + str(page_no) + ' cost: ' + str(time.time()-start_time))

            if self._doc.error_code is None and self._page.error_code is not None:
                if self._page.error_code[0] in [-4, -3, 0]:
                    page_no += 1
                    continue
                else:
                    self._doc.error_code = self._page.error_code
                    break
            self._doc.add_child(self._page)
            page_no += 1

        self.delete_same_image()
        self.delete_header_footer()
        # self.delete_bold_text_duplicate()

    def delete_same_image(self, show=0):
        # 剔除大量重复图片
        md5_dict = {}
        for _md5, image_obj in self.md5_image_obj_list:
            if _md5 in md5_dict.keys():
                md5_dict[_md5] += [image_obj]
            else:
                md5_dict[_md5] = [image_obj]
        cnt_threshold = 10
        delete_obj_list = []
        for _md5 in md5_dict.keys():
            img_list = md5_dict.get(_md5)
            # print('len(md5_dict.get(_md5))', _md5, len(img_list))
            if len(img_list) >= cnt_threshold:
                if show:
                    img_np = bytes2np(img_list[0].content)
                    cv2.namedWindow('delete same img_np', cv2.WINDOW_NORMAL)
                    cv2.imshow('delete same img_np', img_np)
                    cv2.waitKey(0)
                delete_obj_list += img_list
        for page in self._doc.children:
            for obj in delete_obj_list:
                if obj in page.children:
                    page.children.remove(obj)

        if show:
            for page in self._doc.children:
                for obj in page.children:
                    if isinstance(obj, _Image):
                        img_np = bytes2np(obj.content)
                        cv2.imshow('page img_np', img_np)
                        cv2.waitKey(0)

    def delete_header_footer(self):
        sen_dict = {}
        for page in self._doc.children:
            for obj in page.children:
                if isinstance(obj, _Sentence):
                    key = str(obj.content) + ' ' + str(int(obj.y))
                    # print('key', key)
                    if key in sen_dict.keys():
                        sen_dict[key] += [obj]
                    else:
                        sen_dict[key] = [obj]
        for key in sen_dict.keys():
            l = sen_dict.get(key)
            if len(l) >= 2/3 * max(10, len(self._doc.children)):
                for page in self._doc.children:
                    new_children = []
                    for obj in page.children:
                        if isinstance(obj, _Sentence):
                            if obj not in l:
                                new_children.append(obj)
                        else:
                            new_children.append(obj)
                    page.children = new_children
                print('len(l)', len(l), len(self._doc.children))
                print('delete_header_footer l[0]', l[0].content, l[0].y)

        return

    def delete_bold_text_duplicate(self, lt_text_box_list):
        # 拿出所有LTChar
        lt_char_list = []
        for lt_text_box in lt_text_box_list:
            for lt_text_line in lt_text_box:
                for lt_char in lt_text_line:
                    if isinstance(lt_char, LTChar):
                        lt_char_list.append(lt_char)

        # 找出需剔除的
        lt_char_list.sort(key=lambda x: (int(x.bbox[1]), x.bbox[0]))
        delete_list = []
        for i in range(len(lt_char_list)):
            lt_char1 = lt_char_list[i]
            bbox1 = lt_char1.bbox
            # lt_char2 = lt_char_list[i+1]
            # bbox2 = lt_char2.bbox
            if lt_char1 in delete_list:
                continue
            # if lt_char2 in delete_list:
            #     continue
            # if lt_char1.get_text() == lt_char2.get_text() and bbox1[0] <= bbox2[0] <= bbox1[2] <= bbox2[2] \
            #         and int(bbox1[1]) == int(bbox2[1]) and int(bbox1[3]) == int(bbox2[3]) \
            #         and re.search('[\u4e00-\u9fff（）：、，。]', lt_char1.get_text()):
            for j in range(i+1, len(lt_char_list)):
                lt_char2 = lt_char_list[j]
                bbox2 = lt_char2.bbox
                if lt_char2 in delete_list:
                    continue
                if lt_char1.get_text() == lt_char2.get_text() and bbox_iou(bbox1, bbox2) >= 0.3 \
                        and re.search('[\u4e00-\u9fff（）：、，。]', lt_char1.get_text()):
                    # continue
                    delete_list.append(lt_char2)

        # 重新组装
        new_lt_text_box_list = []
        for lt_text_box in lt_text_box_list:
            new_lt_text_box = LTTextBoxHorizontal()
            for lt_text_line in lt_text_box:
                new_lt_text_line = LTTextLine(0.01)
                for lt_char in lt_text_line:
                    if lt_char in delete_list:
                        continue
                    if isinstance(lt_char, LTChar):
                        new_lt_text_line.add(lt_char)
                new_lt_text_box.add(new_lt_text_line)
            new_lt_text_box_list.append(new_lt_text_box)
        return new_lt_text_box_list

    def clean_text(self, _text):
        return re.sub("\s", "", _text)

    def get_text_lines(self, page, page_no):
        lt_line_list = []
        page_plumber = pdfPage(self.doc_pdfplumber, page, page_number=page_no, initial_doctop=self.doc_top)
        self.doc_top += page_plumber.height

        table_finder = TableFinder(page_plumber)
        all_width_zero = True
        for _edge in table_finder.get_edges():
            if _edge.get('linewidth') and _edge.get('linewidth') > 0:
                all_width_zero = False
                break
        for _edge in table_finder.get_edges():
            # print(_edge)
            if _edge.get('linewidth', 0.1) > 0 or all_width_zero:
                lt_line_list.append(LTLine(1, (float(_edge["x0"]), float(_edge["y0"])),
                                           (float(_edge["x1"]), float(_edge["y1"]))))
        log("pdf page_no %s has %s lines" % (str(page_no), str(len(lt_line_list))))
        return lt_line_list

    def get_page_lines(self, layout, page_no, show=0):
        lt_line_list = table_line_pdf(layout, page_no, show)
        return lt_line_list

    def recognize_text(self, layout, page_no, lt_text_list, lt_line_list):
        list_tables, filter_objs, _, connect_textbox_list = self.lt.recognize_table(lt_text_list, lt_line_list, from_pdf=True)
        self._page.in_table_objs = filter_objs

        # print("=======text_len:%d:filter_len:%d"%(len(lt_text_list),len(filter_objs)))

        for table in list_tables:
            _table = _Table(table["table"], table["bbox"])
            # self._page.children.append(_table)
            self._page.add_child(_table)

        list_sentences = ParseUtils.recognize_sentences(lt_text_list, filter_objs,
                                                        layout.bbox, page_no)

        for sentence in list_sentences:
            # print('sentence.text', sentence.text)
            _sen = _Sentence(sentence.text, sentence.bbox)
            self._page.add_child(_sen)
        # pdf对象需反向排序
        self._page.is_reverse = True

        return list_tables

    def is_text_legal(self, lt_text_list, page_no):
        # 无法识别pdf字符编码，整页用ocr
        text_temp = ""
        for _t in lt_text_list:
            text_temp += _t.get_text()

        if re.search('[(]cid:[0-9]+[)]', text_temp):
            log("page_no: " + str(page_no) + " text has cid! try pymupdf...")
            page_image = self.get_page_image(page_no)
            if judge_error_code(page_image):
                self._page.error_code = page_image
            else:
                _image = _Image(page_image[1], page_image[0])
                self._page.add_child(_image)
            return False

        match1 = re.findall(get_garble_code(), text_temp)
        # match2 = re.search('[\u4e00-\u9fa5]', text_temp)
        if len(match1) > 8 and len(text_temp) > 10:
            log("page_no: " + str(page_no) + " garbled code! try pymupdf... " + text_temp[:20])
            page_image = self.get_page_image(page_no)
            if judge_error_code(page_image):
                self._page.error_code = page_image
            else:
                _image = _Image(page_image[1], page_image[0])
                self._page.add_child(_image)
            return False

        return True

    def judge_b_table(self, lt_text_list, table_list, page_no):
        table_h_list = []
        for table in table_list:
            table_h_list.append([table.get('bbox')[1], table.get('bbox')[3]])

        # 先分行
        lt_text_list.sort(key=lambda x: (x.bbox[1], x.bbox[0]))
        lt_text_row_list = []
        current_h = lt_text_list[0].bbox[1]
        row = []
        threshold = 2
        for lt_text in lt_text_list:
            bbox = lt_text.bbox
            if current_h - threshold <= bbox[1] <= current_h + threshold:
                row.append(lt_text)
            else:
                if row:
                    lt_text_row_list.append(row)
                row = [lt_text]
                current_h = lt_text.bbox[1]
        if row:
            lt_text_row_list.append(row)

        # 判断文本中间是否是空格，或一行文本中间有多个
        is_b_table_cnt = 3
        tolerate_cnt = 2
        t_cnt = 0
        row_cnt = 0
        b_table_row_list = []
        all_b_table = []
        for row in lt_text_row_list:
            # 水印行跳过
            if len(row) == 1 and len(row[0].get_text()[:-1]) == 1:
                continue
            # 目录行跳过
            continue_flag = False
            for r in row:
                if re.search('[.·]{7,}', r.get_text()):
                    continue_flag = True
                    break
            if continue_flag:
                continue

            if len(row) == 1:
                text = row[0].get_text()
                bbox = row[0].bbox
                match = re.search('[ ]{3,}', text)
                if match and re.search('[\u4e00-\u9fff]{2,}', text[:match.span()[0]]) \
                        and re.search('[\u4e00-\u9fff]{2,}', text[match.span()[1]:]):
                    row_cnt += 1
                    t_cnt = 0
                    b_table_row_list += row
                else:
                    # 容忍
                    if t_cnt < tolerate_cnt:
                        t_cnt += 1
                        continue
                    if b_table_row_list and row_cnt >= is_b_table_cnt:
                        all_b_table.append(b_table_row_list)
                    row_cnt = 0
                    b_table_row_list = []
            else:
                row_cnt += 1
                t_cnt = 0
                b_table_row_list += row

        if b_table_row_list and row_cnt >= is_b_table_cnt:
            all_b_table.append(b_table_row_list)

        # 对每个可能的b_table判断是否与table相交
        is_b_table_flag = False
        for b_table in all_b_table:
            # 判断在不在有边框表格的范围
            in_flag = False
            for table_h in table_h_list:
                for b in b_table:
                    if min(table_h) <= b.bbox[1] <= max(table_h) or min(table_h) <= b.bbox[3] <= max(table_h):
                        in_flag = True
                        break
                if in_flag:
                    break
            if in_flag:
                is_b_table_flag = False
            else:
                is_b_table_flag = True
                # print('is_b_table_flag True ', [[x.get_text(), x.bbox] for x in b_table])
                # print('table_h_list', table_h_list)
                break
        log("page_no: " + str(page_no) + ' is_b_table_flag ' + str(is_b_table_flag))
        return is_b_table_flag

    def convert_page(self, page, page_no):
        layout = self.get_layout(page, page_no)
        if self._doc.error_code is not None:
            return
        if judge_error_code(layout):
            self._page.error_code = layout
            return

        # 判断该页的对象类型，并存储
        lt_text_list = []
        lt_image_list = []
        for x in layout:
            if isinstance(x, (LTTextBoxHorizontal, LTTextBoxVertical)):
                lt_text_list.append(x)
            if isinstance(x, LTFigure):
                for y in x:
                    if isinstance(y, LTImage):
                        # 小的图忽略
                        if y.width <= 300 and y.height <= 300:
                            continue
                        # 图的width超过layout width，很大可能是水印
                        if y.width > layout.width + 20:
                            continue
                        lt_image_list.append(y)

        # 判断读出来的是乱码，但有图片直接识别
        all_text = ''.join([x.get_text() for x in lt_text_list])
        all_text = re.sub('[\s\d]', '', all_text)
        if len(re.findall(get_garble_code2(), all_text)) >= 3 and len(lt_image_list) >= 1:
            log('嵌入的文字是乱码1: ' + str(all_text[:10]))
            lt_text_list = []
        # print('11111', re.findall(get_traditional_chinese(), all_text))
        if 3 <= len(re.findall(get_traditional_chinese(), all_text)) <= len(all_text) / 2 and len(lt_image_list) >= 1:
            log('嵌入的文字是乱码2: ' + str(all_text[:10]))
            lt_text_list = []

        # 解决重复粗体字问题
        lt_text_list = self.delete_bold_text_duplicate(lt_text_list)

        # 删除水印字
        lt_text_list = self.delete_water_mark(lt_text_list, layout.bbox, 15)
        log("page_no: " + str(page_no) + " len(lt_image_list), len(lt_text_list) " + str(len(lt_image_list)) + " " + str(len(lt_text_list)))

        # 若该页图片数量过多，或无文本，则直接ocr整页识别
        if len(lt_image_list) > 4 or len(lt_text_list) == 0:
            page_image = self.get_page_image(page_no)
            if judge_error_code(page_image):
                self._page.error_code = page_image
            else:
                _image = _Image(page_image[1], page_image[0])
                _image.is_from_pdf = True
                self._page.add_child(_image)

        # 正常读取该页对象
        else:
            # 图表对象
            for image in lt_image_list:
                try:
                    # print("pdf2text LTImage size", page_no, image.width, image.height)
                    image_stream = image.stream.get_data()
                    # 小的图忽略
                    if image.width <= 300 and image.height <= 300:
                        continue
                    # 查看提取的图片高宽，太大则用pdf输出图进行ocr识别
                    img_test = Image.open(io.BytesIO(image_stream))
                    if image.height >= 1000 and image.width >= 1000:
                        page_image = self.get_page_image(page_no)
                        if judge_error_code(page_image):
                            self._page.error_code = page_image
                        else:
                            _image = _Image(page_image[1], page_image[0])
                            _image.is_from_pdf = True
                            self._page.add_child(_image)
                            image_md5 = get_md5_from_bytes(page_image[1])
                            self.md5_image_obj_list.append([image_md5, _image])
                        return
                    # 比较小的图则直接保存用ocr识别
                    else:
                        temp_path = self.unique_type_dir + 'page' + str(page_no) \
                                    + '_lt' + str(lt_image_list.index(image)) + '.jpg'
                        img_test.save(temp_path)
                        with open(temp_path, "rb") as ff:
                            image_stream = ff.read()
                        _image = _Image(image_stream, temp_path, image.bbox)
                        self._page.add_child(_image)
                        image_md5 = get_md5_from_bytes(image_stream)
                        self.md5_image_obj_list.append([image_md5, _image])
                except Exception:
                    log("page_no: " + str(page_no) + " pdfminer read image fail! use pymupdf read image...")
                    traceback.print_exc()

            # pdf对象需反向排序
            self._page.is_reverse = True
            if self.has_init_pdf[3] == 0:
                self.init_package("pdfplumber")

            if not self.is_text_legal(lt_text_list, page_no):
                return

            try:
                lt_line_list = self.get_page_lines(layout, page_no)
            except:
                traceback.print_exc()
                lt_line_list = []
                self._page.error_code = [-13]

            table_list = self.recognize_text(layout, page_no, lt_text_list, lt_line_list)

            # 根据text规律，判断该页是否可能有无边框表格
            if self.judge_b_table(lt_text_list, table_list, page_no):
                page_image = self.get_page_image(page_no)
                if judge_error_code(page_image):
                    self._page.error_code = page_image
                else:
                    _image = _Image(page_image[1], page_image[0])
                    _image.is_from_pdf = True
                    _image.b_table_from_text = True
                    _image.b_table_text_obj_list = lt_text_list
                    _image.b_table_layout_size = (layout.width, layout.height)
                    self._page.add_child(_image)

    def get_layout(self, page, page_no):
        if self.has_init_pdf[0] == 0:
            self.init_package("pdfminer")
        if self._doc.error_code is not None:
            return

        # 获取该页layout
        start_time = time.time()
        try:
            if get_platform() == "Windows":
                layout = pdf_analyze(self.interpreter, page, self.device, page_no)
            else:
                layout = pdf_analyze(self.interpreter, page, self.device, page_no)
        except TimeoutError as e:
            log("page_no: " + str(page_no) + " pdfminer read page time out! " + str(time.time() - start_time))
            layout = [-4]
        except Exception:
            traceback.print_exc()
            log("page_no: " + str(page_no) + " pdfminer read page error! continue...")
            layout = [-3]
        log("page_no: " + str(page_no) + " get_layout cost: " + str(time.time()-start_time))
        return layout

    def get_page_image(self, page_no):
        start_time = time.time()
        try:
            if self.has_init_pdf[1] == 0:
                self.init_package("PyMuPDF")
            if self._doc.error_code is not None:
                return

            # save_dir = self.path.split(".")[-2] + "_" + self.path.split(".")[-1]
            output = self.unique_type_dir + "page" + str(page_no) + ".png"
            page = self.doc_pymupdf.loadPage(page_no)
            rotate = int(0)
            zoom_x = 2.
            zoom_y = 2.
            mat = fitz.Matrix(zoom_x, zoom_y).preRotate(rotate)
            pix = page.getPixmap(matrix=mat, alpha=False)
            pix.writePNG(output)
            # 输出图片resize
            self.resize_image(output)
            with open(output, "rb") as f:
                pdf_image = f.read()
            log("page_no: " + str(page_no) + ' get_page_image cost: ' + str(time.time()-start_time))
            return [output, pdf_image]
        except ValueError as e:
            traceback.print_exc()
            if str(e) == "page not in document":
                log("page_no: " + str(page_no) + " page not in document! continue...")
                return [0]
            elif "encrypted" in str(e):
                log("page_no: " + str(page_no) + " document need password")
                return [-7]
        except RuntimeError as e:
            if "cannot find page" in str(e):
                log("page_no: " + str(page_no) + " page cannot find in document! continue...")
                return [0]
            else:
                traceback.print_exc()
                return [-3]

    def get_all_page_image(self):
        start_time = time.time()
        if self.has_init_pdf[1] == 0:
            self.init_package("PyMuPDF")
        if self._doc.error_code is not None:
            return

        page_count = self.doc_pymupdf.page_count
        for page_no in range(page_count):
            # 限制pdf页数，只取前10页后10页
            if page_count > 20:
                if 10 <= page_no < page_count - 10:
                    continue

            self._page = _Page(None, page_no)
            page_image = self.get_page_image(page_no)
            if judge_error_code(page_image):
                self._page.error_code = page_image
            else:
                _image = _Image(page_image[1], page_image[0])
                self._page.add_child(_image)
            # 报错继续读后面页面
            if self._doc.error_code is None and self._page.error_code is not None:
                continue
            self._doc.add_child(self._page)
        log('get_all_page_image cost: ' + str(time.time()-start_time))

    def connect_table(self, html_list):
        if not html_list:
            return html_list

        # 判断初始条件1
        # 0: 前一页最后一个表格为A，后一页第一个表格为B
        # 1.1: A后无文本(除了页码)，且B前无文本(除了页码)
        # 1.2: B前有文字(可能是页眉，小于60字)，且B的第一行前几个单元格为空，且第一行不为空的单元格有文字较多的格子
        # 1.3: B前有文字(可能是页眉，小于60字)，且B的第一行第一个单元格为空，且有文字的格子数量占所有格子的一半
        # 1.4: B前有文字(可能是页眉，小于60字)，且B的第一行第一个单元格为纯数字序号
        # 1.5: A后有文字(除了页码还有页眉)，且A的后面只有一行且中文不超过15个字
        connect_flag_list = []
        soup_list = []
        for i, h in enumerate(html_list):
            soup = BeautifulSoup(h, 'lxml')
            soup_list.append(soup)
            # 找最后一个表格
            last_table_start, last_table_end = None, None
            match = re.finditer('<table', h)
            for m in match:
                last_table_start = m.span()[0]
            if last_table_start is not None:
                match = re.finditer('</table>', h[last_table_start:])
                for m in match:
                    last_table_end = m.span()[1] + last_table_start
            # 最后一个表格后有无除了页码外的内容
            connect_flag1 = False
            if last_table_end is not None:
                match = re.findall('[^-/第页0-9，,]', re.sub('<div>|</div>', '', h[last_table_end:]))
                # print('match', match.group())
                # if not match or match.group() == '':
                if len(match) == 0:
                    connect_flag1 = True

                # 有页脚
                if not connect_flag1:
                    if len(re.findall('<div>', h[last_table_end:])) <= 1 \
                            and len(re.findall('[\u4e00-\u9fff]', h[last_table_end:])) <= 60:
                        connect_flag1 = True

            # 找第一个表格
            first_table_start, first_table_end = None, None
            match = re.finditer('<table', h)
            for m in match:
                first_table_start = m.span()[0]
                break
            # 第一个表格前有无内容
            connect_flag2 = False
            if first_table_start is not None and first_table_start == 0:
                connect_flag2 = True
            # 有内容但是是页眉
            if not connect_flag2:
                tables = soup.findAll('table')
                if tables:
                    first_table = tables[0]
                    rows = first_table.findAll('tr')
                    if rows:
                        first_row = rows[0]
                        col_text_len_list = [len(x.text) for x in first_row]
                        col_text_list = [x.text for x in first_row]

                        # 文字大于60且第一个为空
                        if not connect_flag2 and len(h[:first_table_start]) <= 60 and col_text_len_list[0] == 0 and max(col_text_len_list) >= 30:
                            connect_flag2 = True
                        # 有文字格子数占一半一下且第一个格子为空
                        if not connect_flag2 and col_text_len_list.count(0) >= len(col_text_len_list) / 2 and col_text_len_list[0] == 0:
                            connect_flag2 = True
                        # 表格前最多只有一行且第一个格子为纯数字
                        if not connect_flag2 and len(col_text_list) > 0 and \
                                len(re.findall('<div>', h[:first_table_start])) <= 0 and \
                                len(re.findall('\d', col_text_list[0])) == len(col_text_list[0]):
                            connect_flag2 = True
                        # if not connect_flag2 and len(re.findall('<div>', h[:first_table_start])) <= 0 and len(re.findall('[\u4e00-\u9fff]', h[:first_table_start])) <= 25:
                        #     connect_flag2 = True

            connect_flag_list.append([i, connect_flag2, connect_flag1])

        print('connect_flag_list', connect_flag_list)

        # 根据条件1合并需连接页码，形成组
        connect_pages_list = []
        if connect_flag_list:
            temp_list = [connect_flag_list[0]]
            for i in range(1, len(connect_flag_list)):
                c = connect_flag_list[i]
                if c[1] and temp_list[-1][2]:
                    temp_list.append(c)
                else:
                    if temp_list:
                        connect_pages_list.append(temp_list)
                        temp_list = [c]
                    # connect_pages_list.append([c])
            if temp_list:
                connect_pages_list.append(temp_list)

        print('connect_pages_list', connect_pages_list)

        # 判断后续条件：判断组内列数是否相同
        connect_pages_list2 = []
        for c_list in connect_pages_list:
            if len(c_list) == 1:
                connect_pages_list2.append(c_list)
            else:
                col_cnt_list = []
                # 单元格可能被复制了，相同的合并当做一列
                merge_col_cnt_list = []
                for c in c_list:
                    soup = soup_list[c[0]]
                    table1 = soup.findAll('table')[-1]
                    table2 = soup.findAll('table')[0]
                    tr1 = table1.findAll('tr')
                    tr2 = table2.findAll('tr')
                    td1 = tr1[-1].findAll('td')
                    td2 = tr2[0].findAll('td')
                    col_cnt_list.append([len(td2), len(td1)])

                    # # 计算合并重复文本格子后的列数
                    # last_text = td1[0].text
                    # merge_td1 = [last_text]
                    # for td in td1:
                    #     if td.text == last_text:
                    #         continue
                    #     else:
                    #         merge_td1.append(td.text)
                    #         last_text = td.text
                    # last_text = td2[0].text
                    # merge_td2 = [last_text]
                    # for td in td2:
                    #     if td.text == last_text:
                    #         continue
                    #     else:
                    #         merge_td2.append(td.text)
                    #         last_text = td.text
                    # merge_col_cnt_list.append([len(merge_td2), len(merge_td1)])

                # 判断
                new_c_list = [c_list[0]]
                # print('col_cnt_list', col_cnt_list)
                for i in range(len(col_cnt_list) - 1):
                    if col_cnt_list[i][1] != col_cnt_list[i + 1][0]:
                            # and merge_col_cnt_list[i][1] != merge_col_cnt_list[i + 1][0]:
                        connect_pages_list2.append(new_c_list)
                        new_c_list = [c_list[i + 1]]
                    else:
                        new_c_list.append(c_list[i + 1])
                if new_c_list:
                    connect_pages_list2.append(new_c_list)

        print('connect_pages_list2', connect_pages_list2)

        # 符合连接条件的拼接表格
        new_html_list = []
        for c_list in connect_pages_list2:
            if len(c_list) == 1:
                new_html_list.append(html_list[c_list[0][0]])
                continue
            new_html = ''
            for c in c_list:
                match = re.finditer('</table>', new_html)
                last_table_index = None
                for m in match:
                    last_table_index = m.span()[0]

                new_html += html_list[c[0]]
                # print('html_list[c[0]]', html_list[c[0]])

                if last_table_index is None:
                    continue

                match = re.finditer('<table border="1">', new_html[last_table_index:])
                first_table_index = None
                for m in match:
                    first_table_index = last_table_index + m.span()[1]
                    break

                if first_table_index is None:
                    continue

                # print('re', re.findall('</table>.*?<table border="1">', new_html[last_table_index:first_table_index]))
                # 非贪婪匹配
                new_html_sub = re.sub('</table>.*?<table border="1">',
                                      '<tr><td>#@#@#</td></tr>',
                                      new_html[last_table_index:first_table_index])
                new_html = new_html[:last_table_index] + new_html_sub + new_html[first_table_index:]
                # print('new_html', new_html)
            # new_html = new_html[:-5]
            # ([-/第页0-9]|<div>|</div>)*
            # 非贪婪匹配
            # match = re.finditer('</table>.*?<table border="1">', new_html)
            # for m in match:
            #     if '#@#@#' in m.group():
            #
            # new_html = re.sub('</table>.*#@#@#.*?<table border="1">',
            #                   '<tr><td>#@#@#</td></tr>',
            #                   new_html)
            # print('new_html', new_html)

            soup = BeautifulSoup(new_html, 'lxml')
            trs = soup.findAll('tr')
            decompose_trs = []
            for i in range(len(trs)):
                if trs[i].get_text() == '#@#@#':
                    td1 = trs[i - 1].findAll('td')
                    td2 = trs[i + 1].findAll('td')
                    if td2[0].get_text() == '':
                        # 解决连续多页是一行表格，该行会被去掉问题
                        find_father = False
                        for father, son in decompose_trs:
                            # print('son', son)
                            # print('td1', trs[i - 1])
                            if father != '' and son == trs[i - 1]:
                                td_father = father.findAll('td')
                                for j in range(len(td_father)):
                                    # print('td_father[j].string3', td_father[j].string)
                                    td_father[j].string = td_father[j].get_text() + td2[j].get_text()
                                    # print('td_father[j].string4', td_father[j].string)
                                    find_father = True
                                    decompose_trs.append([father, trs[i + 1]])
                                break

                        if not find_father:
                            for j in range(len(td1)):
                                # print('td1[j].string1', td1[j].string)
                                td1[j].string = td1[j].get_text() + td2[j].get_text()
                                # print('td1[j].string2', td1[j].string)
                                decompose_trs.append([trs[i - 1], trs[i + 1]])

                        # print('trs[i + 1]', trs[i + 1])
                        # trs[i + 1].decompose()

                    # print('trs[i-1]', trs[i-1])
                    # trs[i].decompose()
                    decompose_trs.append(['', trs[i]])

            # print('decompose_trs', decompose_trs)
            # for father, son in decompose_trs:
            #     print('father', father)
            #     print('son', son)
            # print('len(decompose_trs)', len(decompose_trs))
            for father, son in decompose_trs:
                for tr in trs:
                    if tr == son:
                        tr.decompose()
                        break

            new_html = str(soup)
            new_html_list.append(new_html)

        html_str = ''
        for h in new_html_list:
            html_str += h
        return [html_str]

    def get_html(self):
        if self._doc.error_code is not None:
            return self._doc.error_code

        self.convert()

        if self._doc.error_code is not None:
            return self._doc.error_code
        html = self._doc.get_html(return_list=True)
        # 表格连接
        try:
            html = self.connect_table(html)
        except:
            traceback.print_exc()
            return [-12]
        return html

    def delete_water_mark(self, lt_text_list, page_bbox, times=5):
        # 删除过多重复字句，为水印
        duplicate_dict = {}
        for _obj in lt_text_list:
            t = _obj.get_text()
            if t in duplicate_dict.keys():
                duplicate_dict[t][0] += 1
                duplicate_dict[t][1].append(_obj)
            else:
                duplicate_dict[t] = [1, [_obj]]
        delete_text = []
        for t in duplicate_dict.keys():
            if duplicate_dict[t][0] >= times:
                obj_list = duplicate_dict[t][1]
                obj_list.sort(key=lambda x: x.bbox[3])
                obj_distance_h = abs(obj_list[-1].bbox[3] - obj_list[0].bbox[1])
                obj_list.sort(key=lambda x: x.bbox[2])
                obj_distance_w = abs(obj_list[-1].bbox[2] - obj_list[0].bbox[0])
                if obj_distance_h >= abs(page_bbox[1] - page_bbox[3]) * 0.7 \
                        and obj_distance_w >= abs(page_bbox[0] - page_bbox[2]) * 0.7:
                    delete_text.append(t)
        temp_text_list = []
        for _obj in lt_text_list:
            t = _obj.get_text()
            if t not in delete_text:
                temp_text_list.append(_obj)
        return temp_text_list

    def resize_image(self, img_path, max_size=2000):
        _img = cv2.imread(img_path)
        if _img.shape[0] <= max_size or _img.shape[1] <= max_size:
            return
        else:
            resize_axis = 0 if _img.shape[0] >= _img.shape[1] else 1
            ratio = max_size / _img.shape[resize_axis]
            new_shape = [0, 0]
            new_shape[resize_axis] = max_size
            new_shape[1 - resize_axis] = int(_img.shape[1 - resize_axis] * ratio)
            _img = cv2.resize(_img, (new_shape[1], new_shape[0]))
            cv2.imwrite(img_path, _img)

    def get_single_pdf(self, path, page_no):
        start_time = time.time()
        try:
            pdf_origin = copy.deepcopy(self.doc_pypdf2)
            pdf_new = copy.deepcopy(self.doc_pypdf2_new)
            pdf_new.addPage(pdf_origin.getPage(page_no))

            path_new = path.split(".")[0] + "_split.pdf"
            with open(path_new, "wb") as ff:
                pdf_new.write(ff)

            log("page_no: " + str(page_no) + " get_single_pdf cost: " + str(time.time()-start_time))
            return path_new
        except PyPDF2.utils.PdfReadError as e:
            return [-3]
        except Exception as e:
            log("page_no: " + str(page_no) + " get_single_pdf error!")
            return [-3]


def get_text_font():
    def flags_decomposer(flags):
        """Make font flags human readable."""
        l = []
        if flags & 2 ** 0:
            l.append("superscript")
        if flags & 2 ** 1:
            l.append("italic")
        if flags & 2 ** 2:
            l.append("serifed")
        else:
            l.append("sans")
        if flags & 2 ** 3:
            l.append("monospaced")
        else:
            l.append("proportional")
        if flags & 2 ** 4:
            l.append("bold")
        return ", ".join(l)

    def get_underlined_textLines(page):
        """
        获取某页pdf上的所有下划线文本信息
        :param page: fitz中的一页
        :return: list of tuples,每个tuple都是一个完整的下划线覆盖的整体：[(下划线句, 所在blk_no, 所在line_no), ...]
        """
        paths = page.get_drawings()  # get drawings on the current page

        # 获取该页内所有的height很小的bbox。因为下划线其实大多是这种矩形
        # subselect things we may regard as lines
        lines = []
        for p in paths:
            for item in p["items"]:
                if item[0] == "l":  # an actual line
                    p1, p2 = item[1:]
                    if p1.y == p2.y:
                        lines.append((p1, p2))
                elif item[0] == "re":  # a rectangle: check if height is small
                    r = item[1]
                    if r.width > r.height and r.height <= 2:
                        lines.append((r.tl, r.tr))  # take top left / right points

        # 获取该页的`max_lineheight`，用于下面比较距离使用
        blocks = page.get_text("dict", flags=11)["blocks"]
        max_lineheight = 0
        for b in blocks:
            for l in b["lines"]:
                bbox = fitz.Rect(l["bbox"])
                if bbox.height > max_lineheight:
                    max_lineheight = bbox.height

        underlined_res = []
        # 开始对下划线内容进行查询
        # make a list of words
        words = page.get_text("words")
        # if underlined, the bottom left / right of a word
        # should not be too far away from left / right end of some line:
        for wdx, w in enumerate(words):  # w[4] is the actual word string
            r = fitz.Rect(w[:4])  # first 4 items are the word bbox
            for p1, p2 in lines:  # check distances for start / end points
                if abs(r.bl - p1) <= max_lineheight:  # 当前word的左下满足下划线左下
                    if abs(r.br - p2) <= max_lineheight:  # 当前word的右下满足下划线右下（单个词，无空格）
                        print(f"Word '{w[4]}' is underlined! Its block-line number is {w[-3], w[-2]}")
                        underlined_res.append((w[4], w[-3], w[-2]))  # 分别是(下划线词，所在blk_no，所在line_no)
                        break  # don't check more lines
                    else:  # 继续寻找同line右侧的有缘人，因为有些下划线覆盖的词包含多个词，多个词之间有空格
                        curr_line_num = w[-2]  # line nunmber
                        for right_wdx in range(wdx + 1, len(words), 1):
                            _next_w = words[right_wdx]
                            if _next_w[-2] != curr_line_num:  # 当前遍历到的右侧word已经不是当前行的了（跨行是不行的）
                                break
                            _r_right = fitz.Rect(_next_w[:4])  # 获取当前同行右侧某word的方框4点
                            if abs(_r_right.br - p2) <= max_lineheight:  # 用此word右下点和p2(目标下划线右上点)算距离，距离要小于max_lineheight
                                print(
                                    f"Word '{' '.join([_one_word[4] for _one_word in words[wdx:right_wdx + 1]])}' is underlined! " +
                                    f"Its block-line number is {w[-3], w[-2]}")
                                underlined_res.append(
                                    (' '.join([_one_word[4] for _one_word in words[wdx:right_wdx + 1]]),
                                     w[-3], w[-2])
                                )  # 分别是(下划线词，所在blk_no，所在line_no)
                                break  # don't check more lines
        return underlined_res

    _p = r'C:\Users\Administrator\Desktop\test_pdf\error2-2.pdf'
    doc_pymupdf = read_pymupdf(_p)
    page = doc_pymupdf[0]
    blocks = page.get_text("dict", flags=11)["blocks"]
    for b in blocks:  # iterate through the text blocks
        for l in b["lines"]:  # iterate through the text lines
            for s in l["spans"]:  # iterate through the text spans
                print("")
                font_properties = "Font: '%s' (%s), size %g, color #%06x" % (
                    s["font"],  # font name
                    flags_decomposer(s["flags"]),  # readable font flags
                    s["size"],  # font size
                    s["color"],  # font color
                )
                print(s)
                print("Text: '%s'" % s["text"])  # simple print of text
                print(font_properties)

    get_underlined_textLines(page)


# 以下为现成pdf单页解析接口
class ParseSentence:

    def __init__(self, bbox, fontname, fontsize, _text, _title, title_text, _pattern, title_degree, is_outline,
                 outline_location, page_no):
        (x0, y0, x1, y1) = bbox
        self.x0 = x0
        self.y0 = y0
        self.x1 = x1
        self.y1 = y1
        self.bbox = bbox
        self.fontname = fontname
        self.fontsize = fontsize
        self.text = _text
        self.title = _title
        self.title_text = title_text
        self.groups = _pattern
        self.title_degree = title_degree
        self.is_outline = is_outline
        self.outline_location = outline_location
        self.page_no = page_no

    def __repr__(self):
        return "%s,%s,%s,%d,%s" % (self.text, self.title, self.is_outline, self.outline_location, str(self.bbox))


class ParseUtils:

    @staticmethod
    def getFontinfo(_page):
        for _obj in _page._objs:
            if isinstance(_obj, (LTTextBoxHorizontal, LTTextBoxVertical)):
                for textline in _obj._objs:
                    done = False
                    for lchar in textline._objs:
                        if isinstance(lchar, (LTChar)):
                            _obj.fontname = lchar.fontname
                            _obj.fontsize = lchar.size
                        done = True
                        break
                    if done:
                        break

    @staticmethod
    def recognize_sentences(list_textbox, filter_objs, page_bbox, page_no,
                            remove_space=True, sourceP_LB=True):

        list_textbox.sort(key=lambda x: x.bbox[0])
        list_textbox.sort(key=lambda x: x.bbox[3], reverse=sourceP_LB)

        cluster_textbox = []
        for _textbox in list_textbox:
            if _textbox in filter_objs:
                continue

            _find = False
            for _ct in cluster_textbox:
                if abs(_ct["y"] - _textbox.bbox[1]) < 5:
                    _find = True
                    _ct["textbox"].append(_textbox)
            if not _find:
                cluster_textbox.append({"y": _textbox.bbox[1], "textbox": [_textbox]})

        cluster_textbox.sort(key=lambda x: x["y"], reverse=sourceP_LB)
        list_sentences = []
        for _line in cluster_textbox:
            _textboxs = _line["textbox"]
            _textboxs.sort(key=lambda x: x.bbox[0])

            _linetext = _textboxs[0].get_text()
            for _i in range(1, len(_textboxs)):
                if abs(_textboxs[_i].bbox[0] - _textboxs[_i - 1].bbox[2]) > 60:
                    if _linetext and _linetext[-1] not in (",", "，", "。", ".", "、", "；"):
                        _linetext += "=，="
                _linetext += _textboxs[_i].get_text()

            _linetext = re.sub("[\s\r\n]", "", _linetext)
            _bbox = (_textboxs[0].bbox[0], _textboxs[0].bbox[1],
                     _textboxs[-1].bbox[2], _textboxs[-1].bbox[3])

            _title = None
            _pattern_groups = None
            title_text = ""
            if not _title:
                _groups = ParseUtils.find_title_by_pattern(_textboxs[0].get_text())
                if _groups:
                    _title = _groups[0][0]
                    title_text = _groups[0][1]
                    _pattern_groups = _groups
            if not _title:
                _groups = ParseUtils.find_title_by_pattern(_linetext)
                if _groups:
                    _title = _groups[0][0]
                    title_text = _groups[0][1]
                    _pattern_groups = _groups
            if not _title:
                _title = ParseUtils.rec_incenter(_bbox, page_bbox)

            title_degree = 2
            if not _title:
                _linetext = _linetext.replace("=，=", "，")
            else:
                _linetext = _linetext.replace("=，=", "")
                title_degree = int(_title.split("_")[1])

            # 页码
            if ParseUtils.rec_incenter(_bbox, page_bbox) and re.search("^\d+$", _linetext) is not None:
                continue

            if _linetext == "" or re.search("^，+$", _linetext) is not None:
                continue

            is_outline = False
            outline_location = -1
            _search = re.search("(?P<text>.+?)\.{5,}(?P<nums>\d+)$", _linetext)
            if _search is not None:
                is_outline = True
                _linetext = _search.group("text")
                outline_location = int(_search.group("nums"))

            list_sentences.append(
                ParseSentence(_bbox, _textboxs[-1].__dict__.get("fontname"), _textboxs[-1].__dict__.get("fontsize"),
                              _linetext, _title, title_text, _pattern_groups, title_degree, is_outline,
                              outline_location, page_no))

        # for _sen in list_sentences:
        #     print(_sen.__dict__)

        return list_sentences

    @staticmethod
    def find_title_by_pattern(_text,
                              _pattern="(?P<title_1>(?P<title_1_index_0_0>^第?)(?P<title_1_index_1_1>[一二三四五六七八九十ⅠⅡⅢⅣⅤⅥⅦⅧⅨⅩⅪⅫ]+)(?P<title_1_index_2_0>[、章]))|" \
                                       "(?P<title_3>^(?P<title_3_index_0_1>[ⅠⅡⅢⅣⅤⅥⅦⅧⅨⅩⅪⅫ]+))|" \
                                       "(?P<title_4>^(?P<title_4_index_0_0>第?)(?P<title_4_index_1_1>[一二三四五六七八九十]+)(?P<title_4_index_2_0>[节]))|" \
                                       "(?P<title_11>^(?P<title_11_index_0_0>\d{1,2}[\.．、\s\-]\d{1,2}[\.．、\s\-]\d{1,2}[\.．、\s\-])(?P<title_11_index_1_1>\d{1,2})(?P<title_11_index_2_0>[\.．、\s\-]))|" \
                                       "(?P<title_10>^(?P<title_10_index_0_0>\d{1,2}[\.．、\s\-]\d{1,2}[\.．、\s\-])(?P<title_10_index_1_1>\d{1,2})(?P<title_10_index_2_0>[\.．、\s\-]))|" \
                                       "(?P<title_7>^(?P<title_7_index_0_0>\d{1,2}[\.．、\s\-])(?P<title_7_index_1_1>\d{1,2})(?P<title_7_index_2_0>[\.．、\s\-]))|" \
                                       "(?P<title_6>^(?P<title_6_index_0_1>\d{1,2})(?P<title_6_index_1_0>[\.．、\s\-]))|" \
                                       "(?P<title_15>^(?P<title_15_index_0_0>（?)(?P<title_15_index_1_1>\d{1,2})(?P<title_15_index_2_0>）))|" \
                                       "(?P<title_17>^(?P<title_17_index_0_0>（?)(?P<title_17_index_1_1>[a-zA-Z]+)(?P<title_17_index_2_0>）))|"
                                       "(?P<title_19>^(?P<title_19_index_0_0>（?)(?P<title_19_index_1_1>[一二三四五六七八九十ⅠⅡⅢⅣⅤⅥⅦⅧⅨⅩⅪⅫ]+)(?P<title_19_index_2_0>）))|" \
                              ):
        _se = re.search(_pattern, _text)
        groups = []
        if _se is not None:
            _gd = _se.groupdict()
            for k, v in _gd.items():
                if v is not None:
                    groups.append((k, v))
        if len(groups):
            groups.sort(key=lambda x: x[0])
            return groups
        return None

    @staticmethod
    def rec_incenter(o_bbox, p_bbox):
        p_width = p_bbox[2] - p_bbox[0]
        l_space = (o_bbox[0] - p_bbox[0]) / p_width
        r_space = (p_bbox[2] - o_bbox[2]) / p_width

        if abs((l_space - r_space)) < 0.1 and l_space > 0.2:
            return "title_2"

    @staticmethod
    def is_first_title(_title):
        if _title is None:
            return False
        if re.search("^\d+$", _title) is not None:
            if int(_title) == 1:
                return True
            return False
        if re.search("^[一二三四五六七八九十百]+$", _title) is not None:
            if _title == "一":
                return True
            return False
        if re.search("^[a-z]+$", _title) is not None:
            if _title == "a":
                return True
            return False
        if re.search("^[A-Z]+$", _title) is not None:
            if _title == "A":
                return True
            return False
        if re.search("^[ⅠⅡⅢⅣⅤⅥⅦⅧⅨⅩⅪⅫ]$", _title) is not None:
            if _title == "Ⅰ":
                return True
            return False
        return False

    @staticmethod
    def get_next_title(_title):
        if re.search("^\d+$", _title) is not None:
            return str(int(_title) + 1)
        if re.search("^[一二三四五六七八九十百]+$", _title) is not None:
            _next_title = ParseUtils.make_increase(['一', '二', '三', '四', '五', '六', '七', '八', '九', '十'],
                                                   re.sub("[十百]", '', _title))
            _next_title = list(_next_title)
            _next_title.reverse()
            if _next_title[-1] != "十":
                if len(_next_title) >= 2:
                    _next_title.insert(-1, '十')
            if len(_next_title) >= 4:
                _next_title.insert(-3, '百')
            if _title[0] == "十":
                if _next_title == "十":
                    _next_title = ["二", "十"]
                _next_title.insert(0, "十")
            _next_title = "".join(_next_title)
            return _next_title
        if re.search("^[a-z]+$", _title) is not None:
            _next_title = ParseUtils.make_increase([chr(i + ord('a')) for i in range(26)], _title)
            _next_title = list(_next_title)
            _next_title.reverse()
            return "".join(_next_title)
        if re.search("^[A-Z]+$", _title) is not None:
            _next_title = ParseUtils.make_increase([chr(i + ord('A')) for i in range(26)], _title)
            _next_title = list(_next_title)
            _next_title.reverse()
            return "".join(_next_title)
        if re.search("^[ⅠⅡⅢⅣⅤⅥⅦⅧⅨⅩⅪⅫ]$", _title) is not None:
            _sort = ["Ⅰ", "Ⅱ", "Ⅲ", "Ⅳ", "Ⅴ", "Ⅵ", "Ⅶ", "Ⅷ", "Ⅸ", "Ⅹ", "Ⅺ", "Ⅻ"]
            _index = _sort.index(_title)
            if _index < len(_sort) - 1:
                return _sort[_index + 1]
            return None

    @staticmethod
    def make_increase(_sort, _title, _add=1):
        if len(_title) == 0 and _add == 0:
            return ""
        if len(_title) == 0 and _add == 1:
            return _sort[0]
        _index = _sort.index(_title[-1])
        next_index = (_index + _add) % len(_sort)
        next_chr = _sort[next_index]
        if _index == len(_sort) - 1:
            _add = 1
        else:
            _add = 0
        return next_chr + ParseUtils.make_increase(_sort, _title[:-1], _add)

    @staticmethod
    def rec_serial(_text, o_bbox, p_bbox, fontname, _pattern="(?P<title_1>^[一二三四五六七八九十]+[、])|" \
                                                             "(?P<title_2>^\d+[\.、\s])|" \
                                                             "(?P<title_3>^\d+\.\d+[\.、\s])|" \
                                                             "(?P<title_4>^\d+\.\d+\.\d+[\.、\s])|" \
                                                             "(?P<title_5>^\d+\.\d+\.\d+\.\d+[\.、\s])"):
        # todo :recog the serial of the sentence

        _se = re.search(_pattern, _text)
        if _se is not None:
            _gd = _se.groupdict()
            for k, v in _gd.items():
                if v is not None:
                    return k
        return None


if __name__ == '__main__':
    PDFConvert(r"C:/Users/Administrator/Downloads/1651896704621.pdf", "C:/Users/Administrator/Downloads/1").get_html()