-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpdfImage.py
292 lines (246 loc) · 10.1 KB
/
pdfImage.py
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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
import fitz
from PIL import Image
from utils import *
from whoosh.analysis import Tokenizer, Token
import jieba
from whoosh.index import create_in
from whoosh.fields import *
from whoosh.qparser import QueryParser
import os
import shutil
# import tempfile
LOGO_WIDTH = 398
LOGO_HEIGHT = 137
ix = None
writer = None
class ChineseTokenizer(Tokenizer):
def __call__(self, value, positions=False, chars=False,
keeporiginal=False, removestops=True,
start_pos=0, start_char=0, mode='', **kwargs):
t = Token(positions, chars, removestops=removestops, mode=mode,
**kwargs)
seglist = jieba.cut(value, cut_all=True)
for w in seglist:
t.original = t.text = w
t.boost = 1.0
if positions:
t.pos = start_pos + value.find(w)
if chars:
t.startchar = start_char + value.find(w)
if chars and positions:
t.endchar = start_char + value.find(w) + len(w)
yield t
def ChineseAnalyzer():
return ChineseTokenizer()
def load_pdf(file, dpi=300, skip_page_front=0, skip_page_back=1, skip_block=5, lang='CN'):
"""
Load pdf file, covert to image, description and index it
:param lang:
:param skip_block:
:param skip_page_back:
:param skip_page_front:
:param dpi:
:param file:
:return:
"""
if file.__contains__('\\gradio\\') or file.__contains__('/gradio/'):
print('gradio file')
doc = fitz.open(file)
else:
print('local file')
doc = fitz.open('using_pdfs/' + file)
# load pages
pages = []
for i in range(doc.page_count):
page = doc.load_page(i)
pages.append(page)
# increase dpi to 300
dpi = int(dpi)
scale = dpi / 72 # default dpi of pdf is 72
matrix = fitz.Matrix(scale, scale)
skip_block = int(skip_block)
base_name = os.path.basename(file).split('.')[0]
path_name = f'images/{base_name}'
if os.path.exists(path_name):
shutil.rmtree(path_name)
os.mkdir(path_name)
temp_image_dir = path_name
# temp_image_dir = tempfile.mkdtemp(prefix='images_')
for page in pages[int(skip_page_front):-int(skip_page_back)]: # skip final page
# part1: get image with description in png-pdf
p1dict = page.get_text('dict')
blocks = p1dict['blocks']
page_pix = page.get_pixmap(matrix=matrix, dpi=dpi)
page_im = Image.frombytes("RGB", (page_pix.width, page_pix.height), page_pix.samples)
saved = [] # need to remove if inner a svg image
for i, block in enumerate(blocks[int(skip_block):]): # head and tail of pages should be ignore
if 'image' in block:
# try:
bbox = block['bbox']
# skip image that width=398 and hight=137 -> Typically LOGO
if (bbox[2] - bbox[0])*scale - LOGO_WIDTH <= 10 and (bbox[3] - bbox[1])*scale - LOGO_HEIGHT <= 10:
continue
# Scale the bbox coordinates
cropped = page_im.crop([int(i * scale) for i in bbox])
number = block['number']
file_name = temp_image_dir + f'/{base_name}_imgbmp_{page.number}_{number}'
image_name = file_name + '.png'
# print(image_name)
cropped.save(image_name)
# # Handle text extraction around the image
text_content = get_text_around_image(blocks[skip_block:], i, lang)
title = get_title_of_image(blocks[skip_block:], i, lang)
# print(text_content[:30])
# print(title)
with open(f'{file_name}.txt', 'w', encoding='utf-8') as text_file:
text_file.write(title + '\n' + text_content.replace('\n', ' ')+ f'\nbase name:{base_name}')
saved.append((file_name, [int(i * scale) for i in bbox]))
# except:
# pass
# part2: get image with description in svg-pdf
svg = page.get_svg_image(matrix=fitz.Identity)
image_clips, svg_blocks = parse_page_svg(svg, page.number)
for clip in image_clips:
transform = []
for item in clip[0]:
# print(item, type(item))
if item[0] == '.':
transform.append(float('0' + item))
elif item[0] == '-':
transform.append(float('-0' + item[1:]))
else:
transform.append(float(item))
d = clip[1]
page_id = clip[2]
block_id = clip[3]
matches = re.findall(r'H(\d+\.?\d*)V(\d+\.?\d*)', d)
float_values = [float(value) for value in matches[0]]
box_width = float_values[0]
box_height = float_values[1]
width_scale = transform[0]
height_scale = transform[3]
width_move = transform[4]
height_move = transform[5]
x1 = width_move * scale
y1 = height_move * scale
# x1=347*scale
# y1=587*scale
x2 = x1 + box_width * width_scale * scale
y2 = y1 + box_height * height_scale * scale
if y1 > y2:
y1, y2 = y2, y1
# print(x1, y1, x2, y2)
# 3. 截取并保存图像
# check images in saved, if in or similar, delete it from file system
for i, (file_name, bbox) in enumerate(saved):
if (abs(bbox[0] - x1) < 10\
and abs(bbox[1] - y1) < 10\
and abs(bbox[2] - x2) < 10\
and abs(bbox[3] - y2) < 10) or \
(bbox[0]>x1-10 and bbox[1]>y1-10 and bbox[2]<x2+10 and bbox[3]<y2+10):
os.remove(file_name + '.png')
os.remove(file_name + '.txt')
saved.pop(i)
break
cropped_img = page_im.crop((int(x1), int(y1), int(x2), int(y2)))
file_name = temp_image_dir + f'/{base_name}_imgsvg_{page.number}_{block_id}'
image_name = file_name + '.png'
cropped_img.save(image_name)
# search title and text
text_content = get_svg_text_around_image(svg_blocks, block_id, lang)
title = get_svg_title_around_image(svg_blocks, block_id, lang)
with open(f'{file_name}.txt', 'w', encoding='utf-8') as text_file:
text_file.write(title + '\n' + text_content.replace('\n', ' ') + f'\nbase name:{base_name}')
print(temp_image_dir)
return temp_image_dir
def build_index(file, tmp_dir, lang='CN'):
# Define the schema for the index
if lang == 'CN':
schema = Schema(file_name=ID(stored=True), content=TEXT(analyzer=ChineseAnalyzer(), stored=True))
else:
schema = Schema(file_name=ID(stored=True), content=TEXT(stored=True))
base_name = os.path.basename(file).split('.')[0]
path_name = f'{base_name}'
# index_path = 'indexes/' + path_name + '_index_dir'
index_path = 'indexes/'
# Create an index in a directory
# if os.path.exists(index_path):
# shutil.rmtree(index_path)
# os.mkdir(index_path)
temp_index_dir = index_path
# temp_index_dir = tempfile.mkdtemp(prefix='index_')
global ix
if ix is None:
ix = create_in(temp_index_dir, schema)
global writer
if writer is None:
writer = ix.writer()
# Add documents to the index
# base_name = os.path.basename(file).split('.')[0]
# image_path = f'images{base_name}'
# writer = ix.writer()
for file in os.listdir(tmp_dir):
if file.endswith('.txt'):
file_path = os.path.join(tmp_dir, file)
with open(file_path, 'r', encoding='utf-8') as f:
content = f.read()
writer.add_document(file_name=file[:-4], content=content)
print('==========')
print(content)
print("==========")
writer.commit()
return ix, temp_index_dir
def search(ix, query, lang='CN', k=10):
# Tokenize the query string and join tokens with OR operator
if lang == 'CN':
query_tokens = jieba.cut(query, cut_all=True)
else:
query_tokens = query.split()
or_query = " OR ".join(query_tokens)
parser = QueryParser("content", ix.schema)
myquery = parser.parse(or_query)
with ix.searcher() as searcher:
results = searcher.search(myquery, limit=k)
# Extract and return the file names and descriptions of the top-k hits
results_list = [(hit['file_name'], hit['content'], hit.score) for hit in results]
return results_list
def return_image(file, results_list, tmp_dir):
# base_name = os.path.basename(file).split('.')[0]
# path_name = f'images{base_name}'
titles = []
images = []
for result in results_list:
title = result[1].split('\n')[0].split(':')[-1]
titles.append(title)
images.append(Image.open(tmp_dir + '/' + result[0] + '.png'))
return titles[0], images[0]
# file = 'CA-IS372x-datasheet_cn.pdf'
# file = 'CA-IS3086 datasheet_cn.pdf'
# temp_image_dir = load_pdf(file, lang='CN')
# ix, temp_index_dir = build_index(file, temp_image_dir)
# results_list = search(ix, "波形", lang='CN', k=10)
# ret_img = return_image(file, results_list, temp_image_dir)
# print('title: ' + ret_img[0])
# ret_img[1].show()
# print(os.listdir('using_pdfs'))
# import tqdm
# for file in tqdm.tqdm(os.listdir('using_pdfs')):
# tmd_dir = load_pdf(file)
# ix, tmp_index_dir = build_index('using_pdfs/' + file, tmd_dir)
# #
# writer.commit()
# from whoosh.index import open_dir
# search_ix = open_dir('indexes')
# query = "IF-428x接收端阈值"
# results = search(search_ix, query, lang='CN', k=10)
# for result in results:
# print(result)
#
# from PIL import Image
#
# for result in results:
# image_name = result[0]
# base_name = image_name.split('_img')[0]
# img = Image.open('images/' + base_name + '/' + image_name + '.png')
# image_title = result[1].split('\n')[0].split(':')[1]
# img.show(title=image_title)