对pdf指定区域截图并提取文字
【摘要】
作者：小小明
明镜本清净，明镜亦非台。 本来无一物，何处惹尘埃！
截图PDF指定区域并提取文件
需求：PDF文件结构都一致，对于下图红框区域截图并提取文本
测试pdfplum...
作者：小小明
明镜本清净，明镜亦非台。
本来无一物，何处惹尘埃！
截图PDF指定区域并提取文件
需求：PDF文件结构都一致，对于下图红框区域截图并提取文本
测试pdfplumber库
先试用一下pdfplumber看看能否提取出文本
import pdfplumber
with pdfplumber.open("测试文档.pdf") as p:
page = p.pages[0]
print(page.extract_text())
- 1
- 2
- 3
- 4
运行结果：
Date of Test : 2020-11-05 R
Test Engineer : ? e
s
KAYSER-THREDE Contact Name : WX u
l
00 EVAluation Version: 2.1.7 sample.def ta
1 n
t
0
8
Z0
Y, 6
X,
g] 40 1
n [ . P
o
ati20 ag
r
e e
cel o
ac0 f J
071H 7
-20 .0; Vo = 15 / 2020-11HEAD00ead Acce 822-75
0-40 3.889 m1-0500E2ACleration -HFC
1080 /s; M = 11 RA / CFC SP 1 Res A_202
g]60 60 kg 1000ultant 0_11_
t [ 0
n 5
ulta40 13
s
e _
r0 2
2 5
00
0 F
-200 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 rid
a
time [ms] y
, 6
.1
A1
Analysis Interval: 0 - 1000 [ms] naly.202
Max(61 ms) = 72 g; Min(4.3 ms) = 0.04043 g s0
cHoICn t=. A330m7 (s5(55.64. 6-1 6 -6 .539 m.61s )m; Hs)IC =3 665 =.7 340 g7; ( c5u5.m4 .- A 636m.3s m =s 7);0 H.1I8C g15 = 307 (55.4 - 66.3 ms) is: IA 11:2
T3
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
- 45
试用后发现，pdfplumber提取对这种存在旋转文字的pdf文字提取效果非常糟糕，即使是正常顺序的位置，也出现了交错现象。
通过PyMuPDF实现区域截图和区域文字提取
官方文档：https://pymupdf.readthedocs.io/en/latest/index.html
Github：https://github.com/pymupdf/PyMuPDF
安装：
pip install pymupdf
- 1
截图
先测试截取左下角的部分：
from IPython.display import Image
import fitz
pdfDoc = fitz.open("测试文档.pdf")
page = pdfDoc[0]
mat = fitz.Matrix(1, 1) # 1.5表示放大1.5倍
rect = page.rect
clip = fitz.Rect(0, 0.87*rect.height,
rect.width*0.8, rect.height)
pix = page.getPixmap(matrix=mat, alpha=False, clip=clip)
display(Image(pix.getImageData()))
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- fitz.Matrix(1, 1)的两个参数表示宽度和高度的放大系数，上面的截的图较小可以通过该参数放大
- fitz.Rect有好几种坐标模式，我选择了(x0, y0, x1, y1)这种坐标模式来定位要截取的区域
- page.getPixmap传入放大系数和区域即可获取图片对象，可直接获取图片的数据也可以写入到文件保存起来
再测试截取右上角部分：
clip = fitz.Rect(0.8*rect.width, 0.27*rect.height,
rect.width*0.9, rect.height)
pix = page.getPixmap(matrix=mat.preRotate(-90), alpha=False, clip=clip)
display(Image(pix.getImageData()))
- 1
- 2
- 3
- 4
mat.preRotate(-90)实现了截取区域逆时针旋转90度。
保存图片很简单，只需调用writeImage即可：
pix.writeImage("tmp.png")
- 1
文字提取
通过fitz.Rect要提取文字的区域即可：
a_text = page.getText(clip=clip)
print(a_text)
- 1
- 2
1. Page of J7822-75-HFCA_2020_11_05 13_25
Head Acceleration SP 1 Resultant
11HEAD0000E2ACRA / CFC1000
75 / 2020-11-05
0.0; Vo = 13.889 m/s; M = 1160 kg
Friday, 6.11.2020 11:23
Analysis: IAT
- 1
- 2
- 3
- 4
- 5
- 6
- 7
这段文本提取的效果还不错！
再测试一下左下角部分：
clip = fitz.Rect(0, 0.87*rect.height,
rect.width*0.8, rect.height)
b_text = page.getText(clip=clip)
print(b_text)
- 1
- 2
- 3
- 4
Max(61 ms) = 72 g; Min(4.3 ms) = 0.04043 g
cont. A3ms(56.61 - 59.61 ms) = 65.74 g; cum. A3ms = 70.18 g
HIC = 307 (55.4 - 66.3 ms); HIC36 = 307 (55.4 - 66.3 ms); HIC15 = 307 (55.4 - 66.3 ms)
Analysis Interval: 0 - 1000 [ms]
- 1
- 2
- 3
- 4
文本行顺序处理
文字的行顺序似乎与原始图片的文本顺序不一致。不过我们可以借助pandas自定义排序，还原到一致的顺序。
import pandas as pd
tmp = pd.DataFrame(b_text.splitlines(), columns=["a"])
tmp["b"] = (tmp.a.str[:2]).astype("category")
tmp.b.cat.set_categories(
['An', 're', 'vi', 'Ma', 'co', 'VC', 'ES'], inplace=True)
tmp.sort_values('b', inplace=True)
b_text = '\n'.join(tmp.a.to_list())
print(b_text)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
Analysis Interval: 0 - 1000 [ms]
Max(61 ms) = 72 g; Min(4.3 ms) = 0.04043 g
cont. A3ms(56.61 - 59.61 ms) = 65.74 g; cum. A3ms = 70.18 g
HIC = 307 (55.4 - 66.3 ms); HIC36 = 307 (55.4 - 66.3 ms); HIC15 = 307 (55.4 - 66.3 ms)
- 1
- 2
- 3
- 4
完整代码
import fitz # pip install PyMuPDF
import os
from IPython.display import Image
import pandas as pd
pdf_path = "测试文档.pdf"
if not os.path.exists("imgs"):
os.mkdir("imgs")
result = []
with fitz.open(pdf_path) as pdfDoc:
for i in range(pdfDoc.pageCount):
page_num = i+1
print("--------------", page_num, "--------------")
page = pdfDoc[i]
mat = fitz.Matrix(1.3, 1.3) # 1.5表示放大1.5倍
rect = page.rect
clip = fitz.Rect(0.8*rect.width, 0.27*rect.height,
rect.width*0.9, rect.height) # 想要截取的区域
pix = page.getPixmap(matrix=mat.preRotate(-90),
alpha=False, clip=clip) # 将页面转换为图像
pix.writeImage(f"imgs/{page_num}_a.png")
img1 = pix.getImageData()
display(Image(img1))
a_text = page.getText(clip=clip)
print(a_text)
clip = fitz.Rect(0, 0.87*rect.height,
rect.width*0.8, rect.height)
pix = page.getPixmap(matrix=mat.preRotate(90), alpha=False, clip=clip)
pix.writeImage(f"imgs/{page_num}_b.png")
img2 = pix.getImageData()
display(Image(img2))
b_text = page.getText(clip=clip)
tmp = pd.DataFrame(b_text.splitlines(), columns=["a"])
tmp["b"] = (tmp.a.str[:2]).astype("category")
tmp.b.cat.set_categories(
['An', 're', 'vi', 'Ma', 'co', 'VC', 'ES'], inplace=True)
tmp.sort_values('b', inplace=True)
b_text = '\n'.join(tmp.a.to_list())
print(b_text)
result.append((a_text, b_text))
df = pd.DataFrame(result, columns=["A", "B"])
df.to_excel("result.xlsx", index=False)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
- 45
- 46
前5页运行结果展示：
-------------- 1 --------------
1. Page of J7822-75-HFCA_2020_11_05 13_25
Head Acceleration SP 1 Resultant
11HEAD0000E2ACRA / CFC1000
75 / 2020-11-05
0.0; Vo = 13.889 m/s; M = 1160 kg
Friday, 6.11.2020 11:23
Analysis: IAT
- 1
- 2
- 3
- 4
- 5
- 6
- 7
Analysis Interval: 0 - 1000 [ms]
Max(61 ms) = 72 g; Min(4.3 ms) = 0.04043 g
cont. A3ms(56.61 - 59.61 ms) = 65.74 g; cum. A3ms = 70.18 g
HIC = 307 (55.4 - 66.3 ms); HIC36 = 307 (55.4 - 66.3 ms); HIC15 = 307 (55.4 - 66.3 ms)
- 1
- 2
- 3
- 4
-------------- 2 --------------
2. Page of J7822-75-HFCA_2020_11_05 13_25
Head Acceleration X SP 1
11HEAD0000E2ACXA / CFC1000
75 / 2020-11-05
0.0; Vo = 13.889 m/s; M = 1160 kg
Friday, 6.11.2020 11:23
Analysis: IAT
- 1
- 2
- 3
- 4
- 5
- 6
- 7
Analysis Interval: 0 - 1000 [ms]
Max(65.5 ms) = 8.15 g; Min(52.2 ms) = -7.426 g
- 1
- 2
-------------- 3 --------------
3. Page of J7822-75-HFCA_2020_11_05 13_25
Head Acceleration Y SP 1
11HEAD0000E2ACYA / CFC1000
75 / 2020-11-05
0.0; Vo = 13.889 m/s; M = 1160 kg
Friday, 6.11.2020 11:23
Analysis: IAT
- 1
- 2
- 3
- 4
- 5
- 6
- 7
Analysis Interval: 0 - 1000 [ms]
Max(59.4 ms) = 71.87 g; Min(52 ms) = -9.89 g
- 1
- 2
-------------- 4 --------------
4. Page of J7822-75-HFCA_2020_11_05 13_25
Head Acceleration Z SP 1
11HEAD0000E2ACZA / CFC1000
75 / 2020-11-05
0.0; Vo = 13.889 m/s; M = 1160 kg
Friday, 6.11.2020 11:23
Analysis: IAT
- 1
- 2
- 3
- 4
- 5
- 6
- 7
Analysis Interval: 0 - 1000 [ms]
Max(56.5 ms) = 20.39 g; Min(63.6 ms) = -23.43 g
- 1
- 2
-------------- 5 --------------
5. Page of J7822-75-HFCA_2020_11_05 13_25
Rib Left Upper Displacement Y SP 1
11RIBSLEUPE2DSYC / CFC180
75 / 2020-11-05
0.0; Vo = 13.889 m/s; M = 1160 kg
Friday, 6.11.2020 11:23
Analysis: IAT
- 1
- 2
- 3
- 4
- 5
- 6
- 7
Analysis Interval: 0 - 1000 [ms]
Max(314.8 ms) = 0.2821 mm; Min(52.9 ms) = -33.24 mm
- 1
- 2
…
文章来源: xxmdmst.blog.csdn.net，作者：小小明-代码实体，版权归原作者所有，如需转载，请联系作者。
原文链接：xxmdmst.blog.csdn.net/article/details/111939952
【版权声明】本文为华为云社区用户转载文章，如果您发现本社区中有涉嫌抄袭的内容，欢迎发送邮件进行举报，并提供相关证据，一经查实，本社区将立刻删除涉嫌侵权内容，举报邮箱：
cloudbbs@huaweicloud.com
- 点赞
- 收藏
- 关注作者
评论（0）