Google Colab使用pdfminer提取PDF文本坐标并绘制字符矩形框求助

首先执行以下命令安装依赖：

!pip install pdfminer.six pymupdf

你之前的代码存在三个核心问题：

• 文件打开顺序错误，open(pathh, 'rb')执行时pathh还未赋值，会直接报错
• pdfminer返回的坐标原点是页面左下角，常见绘图库的坐标原点是左上角，未做坐标转换就绘制会出现位置偏移
• 只遍历到了文本块LTTextBox层级，要获取单个字符的坐标需要向下遍历到LTChar对象

完整调整后代码 #

from pdfminer.layout import LAParams, LTTextBox, LTTextLine, LTChar
from pdfminer.pdfpage import PDFPage
from pdfminer.pdfinterp import PDFResourceManager, PDFPageInterpreter
from pdfminer.converter import PDFPageAggregator
import fitz
from google.colab import drive
from PIL import Image

# 挂载谷歌云盘
drive.mount('/content/drive')
# 配置文件路径
pathh = "/content/drive/MyDrive/ICAO.pdf"
output_path = "/content/ICAO_with_boxes.pdf"

# 第一步：用pdfminer提取所有字符/文本块的坐标
rsrcmgr = PDFResourceManager()
laparams = LAParams()
device = PDFPageAggregator(rsrcmgr, laparams=laparams)
interpreter = PDFPageInterpreter(rsrcmgr, device)
fp = open(pathh, 'rb')
pages = list(PDFPage.get_pages(fp))
# 存储每页的文本坐标信息
page_text_coords = []

for page in pages:
    interpreter.process_page(page)
    layout = device.get_result()
    page_width = page.mediabox[2]
    page_height = page.mediabox[3]
    current_page_coords = []
    for lobj in layout:
        if isinstance(lobj, LTTextBox):
            # 遍历到单个字符层级，要文本块级坐标的话可以直接取lobj.bbox
            for line in lobj:
                if isinstance(line, LTTextLine):
                    for char in line:
                        if isinstance(char, LTChar):
                            x0, y0, x1, y1 = char.bbox
                            # 坐标转换：把左下角原点转成左上角原点
                            y0_new = page_height - y1
                            y1_new = page_height - y0
                            current_page_coords.append({
                                "x0": x0, "y0": y0_new,
                                "x1": x1, "y1": y1_new,
                                "text": char.get_text()
                            })
    page_text_coords.append({
        "width": page_width,
        "height": page_height,
        "coords": current_page_coords
    })
fp.close()

# 第二步：用PyMuPDF绘制矩形框
doc = fitz.open(pathh)
for page_idx in range(len(doc)):
    page = doc[page_idx]
    coords_info = page_text_coords[page_idx]
    for item in coords_info["coords"]:
        # 绘制红色矩形框，线宽0.5
        rect = fitz.Rect(item["x0"], item["y0"], item["x1"], item["y1"])
        page.draw_rect(rect, color=(1,0,0), width=0.5)
# 保存带框的PDF
doc.save(output_path)
doc.close()

# 可选：在Colab中直接展示第一页效果
doc = fitz.open(output_path)
page = doc[0]
pix = page.get_pixmap()
img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)
display(img)

效果调整说明 #

• 要绘制文本块级别的框，只需把LTChar层级的遍历逻辑去掉，直接处理LTTextBox的bbox即可
• 矩形框的颜色、线宽可以修改draw_rect方法的color、width参数调整
• 不需要单字符坐标的话可以大幅简化遍历逻辑，运行速度也会更快

内容的提问来源于stack exchange，提问作者engineering Baba