import fitz, os
thisdir = lambda f: os.path.join(os.path.dirname(__file__), f)
thisfile = os.path.abspath(__file__)
outfile = thisfile.replace(".py", ".pdf")
font1 = fitz.Font("helv")
font2 = fitz.Font("tiro")
doc = fitz.open()
page = doc.newPage()
point = fitz.Point(50, 72)
matrix = fitz.Matrix(-20)
wrt1 = fitz.TextWriter(page.rect, color=(0, 0, 1))
wrt2 = fitz.TextWriter(page.rect, color=(1, 0, 0))
_, last = wrt1.append(point, "This text changes color,", font1, 11)
_, last = wrt2.append(last, " font and fontsize", font2, 18)
_, last = wrt1.append(last, " several", font1, 11)
_, last = wrt2.append(last, " times!", font2, 24)
# output both text writers on current page in arbitrary sequence
wrt1.writeText(page, morph=(point, matrix)) # using the same morph parameter
wrt2.writeText(page, morph=(point, matrix)) # also preserves the joint text.
# make a new page
page = doc.newPage()
rect = wrt1.textRect | wrt2.textRect # join rect of blue and red text
# make new rectangle from it, rotated by 90 degrees
nrect = fitz.Rect(
rect.tl, # same top-left, but width and height exchanged
def pyMuPDF_fitz(pdfPath, imagePath, imageName):
startTime_pdf2img = datetime.datetime.now() # 开始时间
print("图片输出路径为：" + imagePath)
print("正在转化，请稍后...")
file_name = os.path.basename(pdfPath) # 获取文件名字
name = file_name.split('.')[0] # 去除后缀，获取名字
pdfDoc = fitz.open(pdfPath)
for pg in range(pdfDoc.pageCount):
page = pdfDoc[pg]
rotate = int(0)
# 每个尺寸的缩放系数为1.3，这将为我们生成分辨率提高2.6的图像。
# 此处若是不做设置，默认图片大小为：792X612, dpi=96
#zoom_x = 1.33333333 # (1.33333333-->1056x816) (2-->1584x1224)
#zoom_y = 1.33333333
# 缩放系数都为2，分辨率提高4倍
#zoom_x = 2 # (1.33333333-->1056x816) (2-->1584x1224)
#zoom_y = 2
#zoom_x = 1 # (1.33333333-->1056x816) (2-->1584x1224)
#zoom_y = 1
zoom_x = 1.111111 # (1.33333333-->1056x816) (2-->1584x1224)
zoom_y = 1.111111
mat = fitz.Matrix(zoom_x, zoom_y).preRotate(rotate)
pix = page.getPixmap(matrix=mat, alpha=False)
if not os.path.exists(imagePath): # 判断存放图片的文件夹是否存在
os.makedirs(imagePath) # 若图片文件夹不存在就创建
# 页码从0开始，水印都加一
pix.writePNG(imagePath + '/' + imageName +
'-%s.JPEG' % str(int(pg) + 1)) # 将图片写入指定的文件夹内
# 为图片添加水印
imageInfo = PIL.Image.open(imagePath + '/' + imageName +
'-%s.JPEG' % str(int(pg) + 1))
# 文字水印
fontOne = ImageFont.truetype("C:\Windows\Fonts\simfang.ttf",
20) # 本地字体文件
draw = ImageDraw.Draw(imageInfo)
#print(imageInfo.size)
# 深灰色fill=(96, 96, 96)
# 开始添加文字水印
draw.text((imageInfo.size[0] // 2 - 25, imageInfo.size[1] // 2 + 445),
u"第%s页" % str(int(pg) + 1),
fill=(0, 0, 0),
font=fontOne)
# imageInfo.show() #展示图片
# 添加图片水印
logo = PIL.Image.open("C:\\logo.png")
w, h = logo.size # 获取图像宽高
logo.thumbnail((800, 800)) # 图像缩小1/2，图像缩放
layer = PIL.Image.new('RGBA', imageInfo.size, (255, 255, 255, 0))
# 添加图片水印
layer.paste(logo, (imageInfo.size[0] - logo.size[0] + 70,
imageInfo.size[1] - logo.size[1] - 20))
imageInfo = PIL.Image.composite(layer, imageInfo, layer)
#imageInfo.paste(logo, (0, 0)) # 将一张图片覆盖到另外一张图片上
imageInfo.save(imagePath + '/' + imageName +
'-%s.JPEG' % str(int(pg) + 1),
quality=1000,
optimize=True,
progressive=True)
#imageInfo.save(imagePath + '/' + imageName + '-%s.JPEG' % str(int(pg) + 1), quality=200, optimize=True)
endTime_pdf2img = datetime.datetime.now() # 结束时间
print('pdf2img转换时间:', (endTime_pdf2img - startTime_pdf2img).seconds)
# 拼接为长图
'''imgs = [Image.open(imagePath + '\\' + fn) for fn in listdir(imagePath) if fn.endswith(".png")] # 打开路径下的所有图片
async def media_loop():
running = True
while running:
originalBookAndPage = False
while not originalBookAndPage:
validBookAndPageOpened = False
while not validBookAndPageOpened:
validBookFound = False
while not validBookFound:
randomBook = random.randint(0, bookCount)
bookToPrint = bookPaths[randomBook]
print(f"Book selected: {bookToPrint}")
try:
book = fitz.open(bookToPrint)
totalPages = book.pageCount
print(f"Book loaded. Page count: {totalPages}")
validBookFound = True
except RuntimeError as e:
print(f"Error with book somewhere. {e}")
# books that don't have valid table of contents page selection process
randomLow = int(totalPages * 0.04)
# randomPageNumber = random.randint(randomLow, totalPages - 8)
randomPageNumber = random.randint(randomLow,
totalPages - randomLow)
if randomPageNumber == 0:
randomPageNumber = 1
pageNumber = randomPageNumber
try:
page = book.loadPage(pageNumber)
print(f"Page loaded: {pageNumber}")
validBookAndPageOpened = True
except RuntimeError as e:
print(f"Error loading page: {pageNumber}. {e}")
try:
zoom = 2
matrix = fitz.Matrix(zoom, zoom)
picOfPage = page.getPixmap(matrix=matrix)
output = f"{bookToPrint}-{pageNumber}.png"
if not book_in_history(bookToPrint, pageNumber):
print(
f"{Fore.RED}\nCOLLISION:\nPage {pageNumber} of {bookToPrint}\n"
)
# if os.path.isfile(output):
# print(
# f"{Fore.RED}\nCOLLISION:\nPage {pageNumber} of {bookToPrint}\n")
else:
originalBookAndPage = True
except:
print(f"{Fore.RED}: Error in checking for collision")
picOfPage.writePNG(output)
status = poststatus(output)
if status:
print(f"{Fore.GREEN}\nSUCCESS\nUploaded: {output}\n")
try:
os.remove(output)
print("File removed succesfully")
except OSError as e:
print(f"Error removing file: {e}")
# print(status.created_at)
else:
print(f"{Fore.RED}\nFailure updating status\n")
print(f"{Fore.BLUE} VALID EVERYTHING")
minutes = 5
time = minutes * 60
print(f"{Fore.YELLOW}{Style.BRIGHT}Sleep for {minutes}m")
await asyncio.sleep(time)
def pdf_ocr(pdf_name, path, method_get_image, words_per_line, ocr_method,
client_id, client_secret):
# 打开pdf
doc = fitz.open(path)
access_token = None
if ocr_method == "online":
# client_id 为官网获取的AK， client_secret 为官网获取的SK
host = 'https://aip.baidubce.com/oauth/2.0/token?grant_type=client_credentials&client_id={}&client_secret={}'.format(
client_id, client_secret)
response = requests.get(host)
if response:
access_token = response.json()["access_token"]
print(access_token)
if if_image_to_pdf_or_hocr:
doc_name = 'output/' + pdf_name[:-4] + "_output.pdf"
else:
doc_name = 'output/' + pdf_name[:-4] + ".docx"
# 正则式提取图片法
if method_get_image == '正则式':
# 使用正则表达式来查找图片
checkXO = r"/Type(?= */XObject)"
checkIM = r"/Subtype(?= */Image)"
# 图片计数
imgcount = 0
lenXREF = doc._getXrefLength()
# 打印PDF的信息
print("文件名:{}, 页数: {}, 对象: {}".format(path, len(doc), lenXREF - 1))
# 遍历每一个对象
for i in range(1, lenXREF):
# 定义对象字符串
text = doc._getXrefString(i)
isXObject = re.search(checkXO, text)
# 使用正则表达式查看是否是图片
isImage = re.search(checkIM, text)
# 如果不是对象也不是图片，则continue
if not isXObject or not isImage:
continue
imgcount += 1
# 根据索引生成图像
pix = fitz.Pixmap(doc, i)
# 根据pdf的路径生成图片的名称
# new_name = path.replace('\\', '_') + "_img{}.png".format(imgcount)
# new_name = new_name.replace(':', '')
# out_image_path = os.path.join(pic_path, new_name)
# # 如果pix.n<5,可以直接存为PNG
# if pix.n < 5:
# pix.writePNG(out_image_path)
# # 否则先转换CMYK
# else:
# pix0 = fitz.Pixmap(fitz.csRGB, pix)
# pix0.writePNG(out_image_path)
# pix0 = None
image = pix.getImageData()
ocr_tesseract(image)
# page = ocr_baidu(image)
# time.sleep(1)
# 释放资源
pix = None
else:
if if_image_to_pdf_or_hocr:
if ocr_method == "online":
raise ValueError('Unsupported filetype for online\
API: {}'.format('pdf'))
return
output_doc = fitz.open()
else:
output_doc = docx.Document()
output_doc.styles['Normal'].font.name = u'等线'
output_doc.styles['Normal']._element.rPr.rFonts\
.set(qn('w:eastAsia'), u'等线')
time_start = time.time()
page_count = doc.pageCount
for pg in range(doc.pageCount):
elapsed = time.time() - time_start
eta = (page_count - pg) * elapsed / pg if pg > 0 else 0
print('[%d/%d] Elapsed: %s, ETA: %s' %
(pg + 1, page_count, fmt_time(elapsed), fmt_time(eta)))
page = doc[pg]
rotate = int(0)
# 每个尺寸的缩放系数为2，这将为我们生成分辨率提高4倍的图像。
zoom_x = 2
zoom_y = 2
trans = fitz.Matrix(zoom_x, zoom_y).preRotate(rotate)
pix = page.getPixmap(matrix=trans, alpha=False)
image = pix.getImageData()
# 这两个标志位解决 pdf 换页有可能不换段的问题
last_para = "" # 上一页的最后一段
last_para_ended = True # 上一页最后一段是否已结束，还是被分页了
if ocr_method == "local":
if if_image_to_pdf_or_hocr:
temp_doc = ocr_tesseract(image)
output_doc.insertPDF(temp_doc, 0, 0)
# output_doc.save(doc_name, incremental=True)
output_doc.save(doc_name)
else:
final_para_ended, paras = ocr_tesseract(
image, words_per_line)
args = [
paras, output_doc, doc_name, last_para,
final_para_ended, last_para_ended
]
last_para, last_para_ended = paras2doc(*args)
else:
final_para_ended, paras = ocr_baidu(image, words_per_line,
access_token)
args = [
paras, output_doc, doc_name, last_para, final_para_ended,
last_para_ended
]
last_para, last_para_ended = paras2doc(*args)
time.sleep(1)
# 释放资源
pix = None
spend = time.time() - start_time
print("Parsing: ", N, " seconds: ", spend)
# Load Documents and meta information
currentDocument = annotations[N]
id = currentDocument['ID']
headerFields = currentDocument["HeaderElements"]
ListElements = currentDocument["ListElements"]
# load document
doc = fitz.open(invoices_path + str(id) + ".pdf")
page = doc[0]
# convert to png
zoom = 4
mat = fitz.Matrix(zoom, zoom)
image = page.getPixmap(matrix = mat, alpha = False)
w = image.width
h = image.height
image.writePNG(png_path + str(id) + ".png")
# Create training data
SegmentImage = np.zeros((height-3, width-1, num_classes), dtype=int)
BoxMasks = np.zeros((height-3, width-1, 2*num_anchors), dtype=int)
BoxCoords = np.zeros((height-3, width-1, 4*num_anchors), dtype=int)
# Fill Seg Mapper with background
for x in range(width-1):
for y in range(height-3):
SegmentImage[y,x] = segMapper("Background")
total_count = missed_count = 0
# iterate over identified words
for word in chain(split_words, standard_words):
wid = word['wid']
total_count += 1
chi_map, eng_map = [], []
# get all bboxs associated with text
for group in word['groups']:
bids, bbox, pg_num = group['bid'], group['bbox'], group['pg']
# obtain image of bbox
bbox = resize(bbox, sf, [eps, 0, eps, 0])
if pg_num in pg_cache:
img = pg_cache[pg_num]
else:
pix = doc[pg_num].getPixmap(matrix=fitz.Matrix(sf, sf))
img = Image.open(io.BytesIO(pix.getPNGData()))
pg_cache[pg_num] = img
block_img = img.crop(bbox)
# perform OCR on bbox
if len(bids) == 1: # single line
chi_map.extend(classify(block_img, chi_detector))
eng_map.extend(classify(block_img, eng_detector))
else: # multi-line
chi_map.extend(classify(block_img, chi_detector_multi))
eng_map.extend(classify(block_img, eng_detector_multi))
# use custom heuristics to obtain text (if high enough probability)
chi_pinyin, chars, char_confs = getChinese(chi_map)
eng_pinyin, nums = getEnglish(eng_map)
guess = refine(chars, char_confs, chi_pinyin, eng_pinyin, bids[0],
letters)
def create_exam_and_insert_QR(
name,
code,
length,
versions,
test,
page_versions,
qr_file,
test_mode=False,
test_folder=None,
):
"""Creates the exam objects and insert the QR codes.
Creates the exams objects from the pdfs stored at sourceVersions.
Then adds the 3 QR codes for each page.
(We create 4 QR codes but only add 3 of them because of the staple side, see below).
Arguments:
name {Str} -- Document Name.
code {Str} -- 6 digit distinguished code for the document.
length {int} -- Length of the document or number of pages.
versions {int} -- Number of version of this Document.
test {int} -- Test number based on the combination we have around (length ^ versions - initial pages) tests .
page_versions {dict} -- (int,int) dictionary representing the version of each page for this test.
qr_file {dict} -- dict(int: dict(int: Str)) Dictionary that has another embedded dictionary for each page.
The embedded dictionary has a string for QR code paths saved for each corner.
Keyword Arguments:
test_mode {bool} -- Boolean elements used for testing, testing case with show the documents. (default: {False})
test_folder {Str} -- String for where to place the generated test files. (default: {None})
Returns:
fitz.Document -- PDF document type returned as the exam, similar to a dictionary with the ge numbers as the keys.
"""
# A (int : fitz.fitz.Document) dictionary that has the page document/path from each source based on page version
version_paths_for_pages = {}
for version_index in range(1, versions + 1):
version_paths_for_pages[version_index] = fitz.open(
"sourceVersions/version{}.pdf".format(version_index))
# Create test pdf as "exam"
exam = fitz.open()
# Insert the relevant page-versions into this pdf.
for page_index in range(1, length + 1):
# Pymupdf starts pagecounts from 0 rather than 1. So offset things.
exam.insertPDF(
version_paths_for_pages[page_versions[page_index]],
from_page=page_index - 1,
to_page=page_index - 1,
start_at=-1,
)
# Get page width and height
page_width = exam[0].bound().width
page_height = exam[0].bound().height
# create a box for the test number near top-centre
rTC = fitz.Rect(page_width // 2 - 50, 20, page_width // 2 + 50, 40)
# put marks at top left/right so students don't write near
# staple or near where client will stamp marks
# create two "do not write" (DNW) rectangles accordingly with TL (top left) and TR (top right)
rDNW_TL = fitz.Rect(15, 15, 90, 90)
rDNW_TR = fitz.Rect(page_width - 90, 15, page_width - 15, 90)
# 70x70 page-corner boxes for the QR codes
# TL: Top Left, TR: Top Right, BL: Bottom Left, BR: Bottom Right
rTL = fitz.Rect(15, 20, 85, 90)
rTR = fitz.Rect(page_width - 85, 20, page_width - 15, 90)
rBL = fitz.Rect(15, page_height - 90, 85, page_height - 20)
rBR = fitz.Rect(page_width - 85, page_height - 90, page_width - 15,
page_height - 20)
for page_index in range(length):
# test/page stamp in top-centre of page
# Rectangle size hacked by hand. TODO = do this more algorithmically
# VALA SAYS: TODO still tands given that the pages are all the same
# size. Will ask what it mean to do it algorithmically
rect = fitz.Rect(page_width // 2 - 40, 20, page_width // 2 + 40, 44)
text = "{}.{}".format(str(test).zfill(4), str(page_index + 1).zfill(2))
insertion_confirmed = exam[page_index].insertTextbox(
rect,
text,
fontsize=18,
color=[0, 0, 0],
fontname="Helvetica",
fontfile=None,
align=1,
)
exam[page_index].drawRect(rect, color=[0, 0, 0])
assert insertion_confirmed > 0
# stamp DNW near staple: even/odd pages different
# Top Left for even pages, Top Right for odd pages
# TODO: Perhaps this process could be improved by putting
# into functions
rDNW = rDNW_TL if page_index % 2 == 0 else rDNW_TR
shape = exam[page_index].newShape()
shape.drawLine(rDNW.top_left, rDNW.top_right)
if page_index % 2 == 0:
shape.drawLine(rDNW.top_right, rDNW.bottom_left)
else:
shape.drawLine(rDNW.top_right, rDNW.bottom_right)
shape.finish(width=0.5, color=[0, 0, 0], fill=[0.75, 0.75, 0.75])
shape.commit()
if page_index % 2 == 0:
# offset by trial-and-error, could be improved
rDNW = rDNW + (19, 19, 19, 19)
else:
rDNW = rDNW + (-19, 19, -19, 19)
mat = fitz.Matrix(45 if page_index % 2 == 0 else -45)
pivot = rDNW.tr / 2 + rDNW.bl / 2
morph = (pivot, mat)
insertion_confirmed = exam[page_index].insertTextbox(
rDNW,
name,
fontsize=8,
fontname="Helvetica",
fontfile=None,
align=1,
morph=morph,
)
# exam[page_index].drawRect(rDNW, morph=morph)
assert (insertion_confirmed >
0), "Text didn't fit: shortname too long? or font issue/bug?"
# Grab the tpv QRcodes for current page and put them on the pdf
# Remember that we only add 3 of the 4 QR codes for each page since
# we always have a corner section for staples and such
qr_code = {}
for corner_index in range(1, 5):
qr_code[corner_index] = fitz.Pixmap(qr_file[page_index +
1][corner_index])
if page_index % 2 == 0:
exam[page_index].insertImage(rTR, pixmap=qr_code[1], overlay=True)
exam[page_index].insertImage(rBR, pixmap=qr_code[4], overlay=True)
exam[page_index].insertImage(rBL, pixmap=qr_code[3], overlay=True)
else:
exam[page_index].insertImage(rTL, pixmap=qr_code[2], overlay=True)
exam[page_index].insertImage(rBL, pixmap=qr_code[3], overlay=True)
exam[page_index].insertImage(rBR, pixmap=qr_code[4], overlay=True)
return exam
pix_2 = fitz.Pixmap(barcode_sign)
page.insertImage(rect_sign, pixmap=pix_2, overlay=True)
page.insertImage(rect_mark, pixmap=pix_1, overlay=True)
doc.save(output_file)
import fitz
for root, dirs, files in walk('D:\\py\\new_CR'):
for f in files:
doc = fitz.open(join(root, f))
width, height = fitz.PaperSize('a4')
totaling = doc.pageCount
for pg in range(totaling):
page = doc[pg]
zoom = int(100)
rotate = int(0)
trans = fitz.Matrix(zoom / 60, zoom / 60).preRotate(rotate)
pm = page.getPixmap(matrix=trans, alpha=False)
lurl = 'D:\\py\\new_CR\\{}.jpg'.format(str(f)[0:12])
pm.writePNG(lurl)
doc.close()
for root, dirs, files in walk('D:\\py\\new_CR'):
for f in files:
if '.jpg' in f:
doc_pdf = fitz.open()
imgdoc = fitz.open(join(root, f))
pdfbytes = imgdoc.convertToPDF()
imgpdf = fitz.open('pdf', pdfbytes)
doc_pdf.insertPDF(imgpdf)
doc_pdf.save('D:\\py\\new_CR\\final\\{}.pdf'.format(str(f)[0:12]))
doc_pdf.close()
def convertPDFPagesToJPG(bookName, bookID):
book = fitz.open(PDFFileDirName + "/" + bookName + ".pdf")
dirToSave = inputsDirName + "/" + bookName
if not os.path.exists(dirToSave): os.makedirs(dirToSave)
for page in book: page.getPixmap(matrix=fitz.Matrix(8, 8)).writeImage(dirToSave + "/" + str(page.number) + ".jpg")
if showPrints: print("Converting PDF pages to JPG pages done")
def make_lessonslists(self, classes, filePath, pdffile):
global boxOfImages
try:
os.makedirs("data")
except:
pass
doc = None
file = pdffile
doc = fitz.open(file)
for i in range(len(doc)):
first_page = doc[i]
image_matrix = fitz.Matrix(fitz.Identity)
image_matrix.preScale(2, 2)
pix = first_page.getPixmap(alpha=False, matrix=image_matrix)
boxOfImages.append(f'{i}.jpg')
pix.writePNG(f'data/{i}.jpg')
NUMBEROFCLASS = 0
for _filename_ in boxOfImages:
img = Image.open(f"{filePath}{_filename_}")
pixMap = img.load()
width, height = img.size
listTemplates = []
boxTime2 = [False]
FIRSTindent, SECONDindent = self.get_indent(pixMap, width, height)
for i in range(FIRSTindent, SECONDindent):
boxTime = []
tok = 0
for j in range(width):
if pixMap[j, i] != (0, 0, 0):
boxTime.append(True)
else:
tok += 1
boxTime.append(False)
# print(boxTime)
if (not all(boxTime2)
and all(boxTime)) or (all(boxTime2)
and not all(boxTime)):
listTemplates.append(i)
boxTime2 = boxTime.copy()
listTemplates.extend([FIRSTindent, SECONDindent])
listTemplates.sort()
try:
os.makedirs("data/data")
except:
pass
def getY(y):
chek = 0
border = set()
variableBorder = None
for i in range(width):
if pixMap[i, y] == (0, 0, 0):
variableBorder = i
elif variableBorder:
border.add(variableBorder)
return sorted(list(border))
BORDERNUM = self.get_num_of_borders(self.get_text_pdf())
for i in range(len(listTemplates) // 2):
try:
if 1:
y0 = listTemplates[i * 2]
y1 = listTemplates[i * 2 + 1]
boxBorder = getY(y0 + 3)
im0 = img.crop((boxBorder[0], y0, boxBorder[2], y1))
for k in range((len(boxBorder) - 3) // BORDERNUM + 1):
# print(boxBorder)
x0 = boxBorder[k * (BORDERNUM + 1) + 2]
x1 = boxBorder[k * (BORDERNUM + 1) + 3 + BORDERNUM]
# print(x0, x1)
im1 = img.crop((x0, y0, x1 + 1, y1))
new_im = Image.new(
'RGB',
(im0.size[0] + im1.size[0], im0.size[1]))
new_im.paste(im0, (0, 0))
new_im.paste(im1, (im0.size[0], 0))
new_im.save('data/data/' +
str(classes[NUMBEROFCLASS]) + '.jpg')
NUMBEROFCLASS += 1
except Exception:
pass
def _get_annots(
self,
annot_image_dir: str = "",
ocr_api: str = "",
zoom: int = 4, # image zoom factor
run_test: bool = False, # get 3 annot and 3 pic at most
):
if not self.doc.has_annots():
return
annot_list = []
annot_count = 0
extracted_pic_count = 0
for page in self.doc.pages():
if run_test and annot_count > 2 and extracted_pic_count > 2:
break
annot_num = 0
word_list = page.getText("words") # list of words on page
word_list.sort(key=lambda w: (w[3], w[0])) # ascending y, then x
for annot in page.annots():
annot_type = annot.type[0]
if annot_type not in ANNOT_TYPES:
continue
page_num = page.number + 1
annot_id = f"annot-{page_num}-{annot_num}"
color = RGB(annot.colors.get("stroke")).to_hex()
height = annot.rect[1] / page.rect[3]
if annot_type == 4: # rectangle
if run_test and extracted_pic_count > 2:
continue
pix = page.get_pixmap(
annots=
False, # TODO donnot display annots, maybe let user customize this?
clip=annot.rect,
matrix=fitz.Matrix(zoom, zoom), # zoom image
)
base_name = self.file_name.replace(" ", "-")
picture_path = os.path.join(annot_image_dir,
f"{base_name}-{annot_id}.png")
pix.writePNG(picture_path)
extracted_pic_count += 1
content = [picture_path]
if ocr_api:
ocr_result = Picture(picture_path).get_ocr_result(
ocr_api)
content.append(ocr_result)
else:
if run_test and annot_count > 2:
continue
content = [annot.info.get("content")]
if annot_type in [8, 9, 10, 11]:
text = self._parse_highlight(annot, word_list)
content.append(text)
annot_list.append({
"type": annot.type[1],
"page": page_num,
"content": content,
"id": annot_id,
"height": height,
"color": color,
})
annot_num += 1
annot_count += 1
return annot_list
class BBox(IText):
'''Boundary box with attribute in fitz.Rect type.'''
# all coordinates are related to un-rotated page in PyMuPDF
# e.g. Matrix(0.0, 1.0, -1.0, 0.0, 842.0, 0.0)
ROTATION_MATRIX = fitz.Matrix(0.0) # rotation angle = 0 degree by default
@classmethod
def set_rotation_matrix(cls, rotation_matrix):
if rotation_matrix and isinstance(rotation_matrix, fitz.Matrix):
cls.ROTATION_MATRIX = rotation_matrix
@classmethod
def pure_rotation_matrix(cls):
'''Pure rotation matrix used for calculating text direction after rotation.'''
a,b,c,d,e,f = cls.ROTATION_MATRIX
return fitz.Matrix(a,b,c,d,0,0)
def __init__(self, raw:dict=None):
''' Initialize BBox and convert to the real (rotation considered) page coordinate system.'''
self.bbox = fitz.Rect()
# NOTE: Any coordinates provided in raw is in original page CS (without considering page rotation).
if raw is None: raw = {}
if 'bbox' in raw:
rect = fitz.Rect(raw['bbox']) * BBox.ROTATION_MATRIX
self.update_bbox(rect)
def __bool__(self):
'''Real object when bbox is defined.'''
return bool(self.bbox)
def __repr__(self): return f'{self.__class__.__name__}({tuple(self.bbox)})'
def get_expand_bbox(self, dt:float):
'''Get expanded bbox with margin dt in both x- and y- direction. Note this method doesn't change its bbox.'''
return self.bbox + (-dt, -dt, dt, dt)
def contains(self, bbox, threshold:float=1.0):
'''Whether given bbox is contained in this instance, with margin considered.'''
# it's not practical to set a general threshold to consider the margin, so two steps:
# - set a coarse but acceptable area threshold,
# - check the length in main direction strictly
if not bbox: return False
# A contains B => A & B = B
intersection = self.bbox & bbox.bbox
factor = round(intersection.getArea()/bbox.bbox.getArea(), 2)
if factor<threshold: return False
# check length
if self.bbox.width >= self.bbox.height:
return self.bbox.width+constants.MINOR_DIST >= bbox.bbox.width
else:
return self.bbox.height+constants.MINOR_DIST >= bbox.bbox.height
def vertically_align_with(self, bbox, factor:float=0.0, text_direction:bool=True):
''' Check whether two boxes have enough intersection in vertical direction, i.e. perpendicular to reading direction.
---
Args:
- bbox: BBox to check with
- factor: threshold of overlap ratio, the larger it is, the higher probability the two bbox-es are aligned.
- text_direction: consider text direction or not. True by default, from left to right if False.
```
+--------------+
| |
+--------------+
L1
+-------------------+
| |
+-------------------+
L2
```
An enough intersection is defined based on the minimum width of two boxes:
```
L1+L2-L>factor*min(L1,L2)
```
'''
if not bbox or not bool(self): return False
# text direction
is_horizontal_text = self.is_horizontal_text if text_direction else True
idx = 0 if is_horizontal_text else 1
L1 = self.bbox[idx+2]-self.bbox[idx]
L2 = bbox.bbox[idx+2]-bbox.bbox[idx]
L = max(self.bbox[idx+2], bbox.bbox[idx+2]) - min(self.bbox[idx], bbox.bbox[idx])
return L1+L2-L>=factor*max(L1,L2)
def horizontally_align_with(self, bbox, factor:float=0.0, text_direction:bool=True):
''' Check whether two boxes have enough intersection in horizontal direction, i.e. along the reading direction.
---
Args:
- bbox: BBox to check with
- factor: threshold of overlap ratio, the larger it is, the higher probability the two bbox-es are aligned.
- text_direction: consider text direction or not. True by default, from left to right if False.
```
+--------------+
| | L1 +--------------------+
+--------------+ | | L2
+--------------------+
```
An enough intersection is defined based on the minimum width of two boxes:
```
L1+L2-L>factor*min(L1,L2)
```
'''
if not bbox or not bool(self): return False
# text direction
is_horizontal_text = self.is_horizontal_text if text_direction else True
idx = 1 if is_horizontal_text else 0
L1 = self.bbox[idx+2]-self.bbox[idx]
L2 = bbox.bbox[idx+2]-bbox.bbox[idx]
L = max(self.bbox[idx+2], bbox.bbox[idx+2]) - min(self.bbox[idx], bbox.bbox[idx])
return L1+L2-L>=factor*max(L1,L2)
def copy(self):
'''make a deep copy.'''
return copy.deepcopy(self)
def update_bbox(self, rect):
'''Update current bbox to specified `rect`.
---
Args:
- rect: fitz.rect or raw bbox like (x0, y0, x1, y1) in real page CS (with rotation considered).
'''
self.bbox = fitz.Rect([round(x,1) for x in rect])
return self
def union_bbox(self, bbox):
'''Update current bbox to the union with specified `rect`.
---
Args:
- bbox: BBox, the target to get union
'''
return self.update_bbox(self.bbox | bbox.bbox)
def compare(self, bbox, threshold=0.9):
'''Whether has same type and bbox.'''
if not isinstance(bbox, self.__class__):
return False, f'Inconsistent type: {self.__class__.__name__} v.s. {bbox.__class__.__name__} (expected)'
if not get_main_bbox(self.bbox, bbox.bbox, threshold):
return False, f'Inconsistent bbox: {self.bbox} v.s. {bbox.bbox}(expected)'
return True, ''
def store(self):
'''Store in json format.'''
return { 'bbox': tuple([x for x in self.bbox]) }
def plot(self, page, stroke:tuple=(0,0,0), width:float=0.5, fill:tuple=None):
'''Plot bbox in PDF page.'''
page.drawRect(self.bbox, color=stroke, fill=fill, width=width, overlay=False)
doc = fitz.open() # empty new PDF
page = doc.newPage() # create page (A4)
img = page.newShape() # create shape
# =============================================================================
# pencil 1
# =============================================================================
penheight = 100 # thickness of pencil
pentip = fitz.Point(100, 150) # first pencil tip here
pencil(img, pentip, penheight, True) # pencil points left
# =============================================================================
# pencil 2
# =============================================================================
penheight = 20 # now a smaller one
pentip = fitz.Point(100, 250) # new pencil tip
pencil(img, pentip, penheight, False) # this one points right
pentip.x += 10 # insert a little distance
text = """Like the ReportLab User Guide does,\nyou may want to use this image, to\nemphasize content, e.g. cautionary\nremarks, notes, examples, etc."""
page.insertText(pentip, text) # insert explanatory text
# =============================================================================
# pencil 3
# =============================================================================
# yet another pencil, which we will morph around its tip
mat = fitz.Matrix(-150) * fitz.Matrix(0.5, 0.5,
1) # morphing: rotate & shear
pentip = fitz.Point(300, 400)
# instead of another thickness (40) we could have used a scale matrix
pencil(img, pentip, 40, True, morph=(pentip, mat))
img.commit()
doc.save("pencil.pdf")
def pdf_filter():
global FILTER_STATUS
# 等待被触发
while WINDOW_STATUS:
if FILTER_STATUS:
FILTER_STATUS = False
# 取消Button功能，触发进度提示
bt['state'] = 'disabled'
progress = 0
var_01.set('正在处理，请稍候...')
var_10.set(f'处理进度：{progress}%')
# pdf转png
try:
input_pdf = filedialog.askopenfile(title='选择pdf文档').name
except:
close_window()
input_doc = fitz.open(input_pdf)
toc = input_doc.getToC() # 获取待处理pdf的目录
page_sum = input_doc.pageCount
for i in range(0, page_sum):
page = input_doc[i]
zoom = 100 * RESOLUTION # 缩放
rotate = 0 # 无转动
trans = fitz.Matrix(zoom/100.0, zoom/100.0).preRotate(rotate)
pm = page.getPixmap(matrix=trans, alpha=False)
if(i+1<10):
page_num = '000' + str(i+1)
elif(i+1<100):
page_num = '00' + str(i+1)
elif(i+1<1000):
page_num = '0' + str(i+1)
else:
page_num = str(i+1)
pm.writePNG('pdf2png/%s.png' % page_num)
progress = int(40 * (i/page_sum))
var_10.set(f'处理进度：{progress}%')
input_doc.close()
# png图片处理
m = page_sum
path = ('pdf2png/')
f = listdir(path)
for i in f:
if i=='0000.png':
continue
img = Image.open(path+i)
if FILTER_MODE == '反色':
inv_img = PIL.ImageOps.invert(img)
elif FILTER_MODE == '灰度':
inv_img = PIL.ImageOps.grayscale(img)
elif FILTER_MODE == '去边':
inv_img = PIL.ImageOps.crop(img, border=10)
elif FILTER_MODE == '增强':
inv_img = PIL.ImageOps.autocontrast(img, cutoff=10)
else: # FILTER_MODE == '跳阶'
inv_img = PIL.ImageOps.posterize(img, 2)
inv_img.save(path+i)
m -= 1
progress = int(40 + 30 * (page_sum-m)/page_sum)
var_10.set(f'处理进度：{progress}%')
# png转pdf
n = page_sum
output_doc = fitz.open()
for img in sorted(glob.glob('pdf2png/*')):
imgdoc = fitz.open(img)
pdfbytes = imgdoc.convertToPDF()
imgpdf = fitz.open('pdf', pdfbytes)
output_doc.insertPDF(imgpdf)
n -= 1
progress = int(70 + 30 * (page_sum-n)/page_sum)
var_10.set(f'处理进度：{progress}%')
var_01.set('处理完成！')
#output_path = filedialog.askdirectory(title='请选择保存位置')+'/'
output_path = filedialog.asksaveasfilename(title='另存为输出pdf文档', defaultextension='.pdf', initialfile=f'{FILTER_MODE} - '+input_pdf.split('/')[-1], filetypes=[('PDF','*.pdf')])
output_doc.setToC(toc) # 将原pdf的目录加入新pdf中
output_doc.save(output_path)
output_doc.close()
close_window()
else:
sleep(0.3)
def image_extract(path):
keywords = "........"
for file in os.listdir(path):
try:
if file[-4:] == ".pdf":
mkpath = 'E:\\项目\\试运行\\201808\\提取图片\\'
mkpath = mkpath + file[:-4] + '\\'
mkdir(mkpath)
doc = fitz.open(os.path.join(path, file))
page_count = doc.pageCount
picname_num = 0
picname_list = []
for i in range(10):
page = doc.loadPage(i)
page_text = page.getText()
if keywords in page_text:
pattern = re.compile(r'图表 [0-9][^\.]*')
picname_list1 = pattern.findall(page_text)
picname_list.extend(picname_list1)
# print(picname_list)
for i in range(2, page_count):
# page = doc[i]
# links = page.getLinks()
page = doc.loadPage(i)
page_text = page.getText()
pic_1 = page.searchFor("图表")
data_1 = page.searchFor("资料来源：")
if len(data_1) == 0:
data_1 = page.searchFor("来源：")
# print('未处理前---第', i, '页--图表的个数为', len(pic_1))
# print('未处理前---第', i, '页--数据来源的个数为', len(data_1))
# 处理重复数据
pic_2 = deal_repeat(pic_1)
data = deal_repeat(data_1)
# print('未处理正文重复----第', i, '页--图表的个数为', len(pic_2))
# print('未处理正文重复--图表坐标', pic_2)
# print('未处理正文重复--资料来源坐标', data)
# print('第', i, '页--数据来源的个数为', len(data))
# 处理正文出现的重复数据
pic = deal_reContext(pic_2, data)
# print('处理完正文重复----第', i, '页--图表的个数为', len(pic))
# print('处理完正文重复----图表坐标', pic)
# print('处理完正文重复----资料来源坐标', data)
# print('第', i, '页--数据来源的个数为', len(data))
# if i == 7:
# print('未处理表格的坐标', pic_1)
# print('处理后表格的坐标', pic)
# print('未处理资料来源的坐标', data_1)
# print('处理后资料来源的坐标', data)
if len(pic) > len(data): # 图表数多于资料来源
pic = pic[:len(data) - len(pic)]
elif len(pic) < len(data): # 资料来源多于图表数
for i in range(len(data) - len(pic)):
pic.insert(0, fitz.Rect(0, 0, 0, 0))
if len(pic) != 0 and len(data) != 0 and len(pic) == len(data):
mat = fitz.Matrix(3, 3) # 缩放
page_ = page.rect # 页面大小
page_length = page_.x1 # 页面长
page_width = page_.y1 # 页面宽
# words = page.getTextWords() #获取页面文字
# print(words)
pic1 = [[]]
data1 = [[]]
length = len(pic) # 图表数
for i in range(length): # 将图表按行分组
if i < (length - 1):
if pic[i].y0 == pic[i + 1].y0:
pic1[-1].append(pic[i])
data1[-1].append(data[i])
else:
pic1[-1].append(pic[i])
pic1.append([])
data1[-1].append(data[i])
data1.append([])
pic1[-1].append(pic[i])
data1[-1].append(data[i])
picgroup_num = len(pic1) # 图片组数
for i in range(picgroup_num): # 按组处理图片
for j in range(len(pic1[i])):
if j < len(pic1[i]) - 1:
clip = fitz.Rect(data1[i][j].x0 - 5, pic1[i][j].y0, data1[i][j + 1].x0 - 18,
data1[i][j].y1)
pix = page.getPixmap(matrix=mat, clip=clip, alpha=False)
# print(pix)
# 预处理字符串中的非法字符
deal_name = validateTitle(picname_list[picname_num])
fn = deal_name + ".png"
pix.writePNG(os.path.join(mkpath, fn))
picname_num = picname_num + 1
else:
clip = fitz.Rect(data1[i][j].x0 - 5, pic1[i][j].y0, page_length - 49,
data1[i][j].y1)
pix = page.getPixmap(matrix=mat, clip=clip, alpha=False)
deal_name = validateTitle(picname_list[picname_num])
fn = deal_name + ".png"
pix.writePNG(os.path.join(mkpath, fn))
picname_num = picname_num + 1
doc.close()
shutil.copy(os.path.join(path, file), os.path.join(mkpath, file))
except:
print(file)
mkdir(path+'\\未处理完成')
shutil.move(mkpath, path+'\\未处理完成')
continue
class Element(IText):
'''Boundary box with attribute in fitz.Rect type.'''
# all coordinates are related to un-rotated page in PyMuPDF
# e.g. Matrix(0.0, 1.0, -1.0, 0.0, 842.0, 0.0)
ROTATION_MATRIX = fitz.Matrix(0.0) # rotation angle = 0 degree by default
@classmethod
def set_rotation_matrix(cls, rotation_matrix):
"""Set global rotation matrix.
Args:
Rotation_matrix (fitz.Matrix): target matrix
"""
if rotation_matrix and isinstance(rotation_matrix, fitz.Matrix):
cls.ROTATION_MATRIX = rotation_matrix
@classmethod
def pure_rotation_matrix(cls):
'''Pure rotation matrix used for calculating text direction after rotation.'''
a, b, c, d, e, f = cls.ROTATION_MATRIX
return fitz.Matrix(a, b, c, d, 0, 0)
def __init__(self, raw: dict = None, parent=None):
''' Initialize Element and convert to the real (rotation considered) page coordinate system.'''
self.bbox = fitz.Rect()
self._parent = parent # type: Element
# NOTE: Any coordinates provided in raw is in original page CS (without considering page rotation).
if 'bbox' in (raw or {}):
rect = fitz.Rect(raw['bbox']) * Element.ROTATION_MATRIX
self.update_bbox(rect)
def __bool__(self):
'''Real object when bbox is defined.'''
return bool(self.bbox)
def __repr__(self):
return f'{self.__class__.__name__}({tuple(self.bbox)})'
# ------------------------------------------------
# parent element
# ------------------------------------------------
@property
def parent(self):
return self._parent
@parent.setter
def parent(self, parent):
self._parent = parent
# ------------------------------------------------
# bbox operations
# ------------------------------------------------
def copy(self):
'''make a deep copy.'''
# NOTE: can't serialize data because parent is an Object,
# so set it None in advance.
parent, self.parent = self._parent, None
obj = copy.deepcopy(self)
self._parent = parent # set back parent
return obj
def get_expand_bbox(self, dt: float):
"""Get expanded bbox with margin in both x- and y- direction.
Args:
dt (float): Expanding margin.
Returns:
fitz.Rect: Expanded bbox.
.. note::
This method creates a new bbox, rather than changing the bbox of itself.
"""
return self.bbox + (-dt, -dt, dt, dt)
def update_bbox(self, rect):
'''Update current bbox to specified ``rect``.
Args:
rect (fitz.Rect or list): bbox-like ``(x0, y0, x1, y1)`` in real page CS (with rotation considered).
'''
self.bbox = fitz.Rect([round(x, 1) for x in rect])
return self
def union_bbox(self, e):
"""Update current bbox to the union with specified Element.
Args:
e (Element): The target to get union
Returns:
Element: self
"""
return self.update_bbox(self.bbox | e.bbox)
# --------------------------------------------
# location relationship to other Element instance
# --------------------------------------------
def contains(self, e, threshold: float = 1.0):
"""Whether given element is contained in this instance, with margin considered.
Args:
e (Element): Target element
threshold (float, optional): Intersection rate. Defaults to 1.0. The larger, the stricter.
Returns:
bool: [description]
"""
# NOTE the case bool(e)=True but e.bbox.get_area()=0
S = e.bbox.get_area()
if not S: return False
# it's not practical to set a general threshold to consider the margin, so two steps:
# - set a coarse but acceptable area threshold,
# - check the length in main direction strictly
# A contains B => A & B = B
intersection = self.bbox & e.bbox
factor = round(intersection.get_area() / e.bbox.get_area(), 2)
if factor < threshold: return False
# check length
if self.bbox.width >= self.bbox.height:
return self.bbox.width + constants.MINOR_DIST >= e.bbox.width
else:
return self.bbox.height + constants.MINOR_DIST >= e.bbox.height
def get_main_bbox(self, e, threshold: float = 0.95):
"""If the intersection with ``e`` exceeds the threshold, return the union of these two elements; else return None.
Args:
e (Element): Target element.
threshold (float, optional): Intersection rate. Defaults to 0.95.
Returns:
fitz.Rect: Union bbox or None.
"""
bbox_1 = self.bbox
bbox_2 = e.bbox if hasattr(e, 'bbox') else fitz.Rect(e)
# areas
b = bbox_1 & bbox_2
if not b: return None # no intersection
a1, a2, a = bbox_1.get_area(), bbox_2.get_area(), b.get_area()
# Note: if bbox_1 and bbox_2 intersects with only an edge, b is not empty but b.get_area()=0
# so give a small value when they're intersected but the area is zero
factor = a / min(a1, a2) if a else 1e-6
return bbox_1 | bbox_2 if factor >= threshold else None
def vertically_align_with(self,
e,
factor: float = 0.0,
text_direction: bool = True):
'''Check whether two Element instances have enough intersection in vertical direction, i.e. perpendicular to reading direction.
Args:
e (Element): Object to check with
factor (float, optional): Threshold of overlap ratio, the larger it is, the higher probability the two bbox-es are aligned.
text_direction (bool, optional): Consider text direction or not. True by default, from left to right if False.
Returns:
bool: [description]
Examples::
+--------------+
| |
+--------------+
L1
+-------------------+
| |
+-------------------+
L2
An enough intersection is defined based on the minimum width of two boxes::
L1+L2-L>factor*min(L1,L2)
'''
if not e or not bool(self): return False
# text direction
idx = 1 if text_direction and self.is_vertical_text else 0
L1 = self.bbox[idx + 2] - self.bbox[idx]
L2 = e.bbox[idx + 2] - e.bbox[idx]
L = max(self.bbox[idx + 2], e.bbox[idx + 2]) - min(
self.bbox[idx], e.bbox[idx])
eps = 1e-3 # tolerent
return L1 + L2 - L + eps >= factor * min(L1, L2)
def horizontally_align_with(self,
e,
factor: float = 0.0,
text_direction: bool = True):
'''Check whether two Element instances have enough intersection in horizontal direction, i.e. along the reading direction.
Args:
e (Element): Element to check with
factor (float, optional): threshold of overlap ratio, the larger it is, the higher probability the two bbox-es are aligned.
text_direction (bool, optional): consider text direction or not. True by default, from left to right if False.
Examples::
+--------------+
| | L1 +--------------------+
+--------------+ | | L2
+--------------------+
An enough intersection is defined based on the minimum width of two boxes::
L1+L2-L>factor*min(L1,L2)
'''
if not e or not bool(self): return False
# text direction
idx = 0 if text_direction and self.is_vertical_text else 1
L1 = self.bbox[idx + 2] - self.bbox[idx]
L2 = e.bbox[idx + 2] - e.bbox[idx]
L = max(self.bbox[idx + 2], e.bbox[idx + 2]) - min(
self.bbox[idx], e.bbox[idx])
eps = 1e-3 # tolerent
return L1 + L2 - L + eps >= factor * min(L1, L2)
def in_same_row(self, e):
"""Check whether in same row/line with specified Element instance. With text direction considered.
Taking horizontal text as an example:
* yes: the bottom edge of each box is lower than the centerline of the other one;
* otherwise, not in same row.
Args:
e (Element): Target object.
Returns:
bool: [description]
.. note::
The difference to method ``horizontally_align_with``: they may not in same line, though
aligned horizontally.
"""
if not e or self.is_horizontal_text != e.is_horizontal_text:
return False
# normal reading direction by default
idx = 1 if self.is_horizontal_text else 0
c1 = (self.bbox[idx] + self.bbox[idx + 2]) / 2.0
c2 = (e.bbox[idx] + e.bbox[idx + 2]) / 2.0
res = c1 <= e.bbox[idx + 2] and c2 <= self.bbox[
idx + 2] # Note y direction under PyMuPDF context
return res
# ------------------------------------------------
# others
# ------------------------------------------------
def store(self):
'''Store properties in raw dict.'''
return {'bbox': tuple([x for x in self.bbox])}
def plot(self,
page,
stroke: tuple = (0, 0, 0),
width: float = 0.5,
fill: tuple = None,
dashes: str = None):
'''Plot bbox in PDF page for debug purpose.'''
page.draw_rect(self.bbox,
color=stroke,
fill=fill,
width=width,
dashes=dashes,
overlay=False,
fill_opacity=0.5)
# Define start / end points of x axis that we want to use as 0 and 2*pi.
# They may be oriented in any way.
#--------------------------------------------------------------------------
pb = fitz.Point(200, 200) # begin, treated as (0, 0)
pe = fitz.Point(400, 100) # end, treated as (2*pi, 0)
alfa = img.horizontal_angle(pb, pe) # connection angle towards x-axis
rad = abs(pe - pb) # distance of these points
pe1 = pb + (rad, 0) # make corresp. horizontal end point
# =============================================================================
# first draw a rectangle in which the functions graphs will later appear
# =============================================================================
f = abs(pe - pb) * 0.5 / math.pi # represents 1 unit
rect = fitz.Rect(pb.x - 5, pb.y - f - 5, pe1.x + 5, pb.y + f + 5)
img.drawRect(rect) # draw it
morph = (pb, fitz.Matrix(math.degrees(-alfa)))
img.finish(fill=yellow, morph=morph) # rotate it around begin point
# =============================================================================
# get all points for the sine function
# =============================================================================
pntsin = bsinPoints(pb, pe1) # only horizontal axis supported
# therefore need rotate result points by angle alfa afterwards
points = rot_points(pntsin, pb, alfa)
for i in (0, 3, 6, 9): # draw all 4 function segments
img.drawBezier(points[i], points[i + 1], points[i + 2], points[i + 3])
img.finish(color=red, width=w, closePath=False)
# =============================================================================
def pure_rotation_matrix(cls):
'''Pure rotation matrix used for calculating text direction after rotation.'''
a, b, c, d, e, f = cls.ROTATION_MATRIX
return fitz.Matrix(a, b, c, d, 0, 0)
def getImg(pg_num, bbox):
bbox_resize = resize(bbox, sf, eps)
pix = doc[pg_num].getPixmap(matrix=fitz.Matrix(sf, sf))
img = Image.open(io.BytesIO(pix.getPNGData()))
block_img = img.crop(bbox_resize)
return block_img
"""
if __name__ == "__main__":
green = getColor("limegreen")
red = getColor("red2")
doc = fitz.open()
p = doc.newPage()
img = p.newShape()
r = fitz.Rect(100, 100, 200, 200)
heart(img, r, red)
img.commit()
p.setCropBox(r + (10, 10, -10, -15))
p = doc.newPage()
img = p.newShape()
pnt = r.tl + (r.br - r.tl) * 0.5
clover(img, r, green, morph=(pnt, fitz.Matrix(45)))
img.commit()
p.setCropBox(r + (5, 5, -5, -5))
p = doc.newPage()
img = p.newShape()
diamond(img, r, red)
img.commit()
p.setCropBox(r)
p = doc.newPage()
img = p.newShape()
pnt = r.tl + (r.br - r.tl) * 0.5
caro(img, r, red, morph=(pnt, fitz.Matrix(45)))
img.commit()
p.setCropBox(r + (10, 10, -10, -10))
# https://www.jianshu.com/p/2abe38044446
# pip install PyMuPDF
# 将本目录下所有PDF文件转为png文件（第8行）
import fitz
import sys
import glob
pdffile = glob.glob("*.pdf")
pngfile = []
for f in pdffile:
pngfile.append(f.rstrip("pdf"))
for i in range(len(pdffile)):
doc = fitz.open(pdffile[i])
page = doc[0]
zoom = int(100)
rotate = int(0)
trans = fitz.Matrix(zoom / 100.0, zoom / 100.0).preRotate(rotate)
pm = page.getPixmap(matrix=trans, alpha=False)
pm.writePNG(pngfile[i] + "png")
def PdftoImage_totext(file_path):
if os.path.exists(file_path):
re1 = r'[\w\W]+(\d{4})[\w\W]+(G\d{5})号$'
file_path_1 = file_path[:file_path.rfind(".")]
pdf = fitz.open(file_path)
if pdf.pageCount > 2:
png_name = file_path_1 + ".png"
png_name2 = file_path_1 + "_jie.png"
if os.path.exists(png_name):
os.remove(png_name)
if os.path.exists(png_name2):
os.remove(png_name2)
rotate = int(0)
# 每个尺寸的缩放系数为1.3，这将为我们生成分辨率提高2.6的图像。
# 此处若是不做设置，默认图片大小为：792X612, dpi=96
zoom_x = 1.33333333 # (1.33333333-->1056x816) (2-->1584x1224)
zoom_y = 1.33333333
mat = fitz.Matrix(zoom_x, zoom_y).preRotate(rotate)
png = pdf[2].getPixmap(matrix=mat, alpha=False)
png.writePNG(png_name)
pngg = cv2.imread(png_name)
kuan = pngg.shape[0]
chang = int(pngg.shape[1] * 0.35)
print(pngg.shape)
cropped = pngg[0:chang, 0:kuan]
cv2.imwrite(png_name2, cropped)
baidu = BaiduApi()
text = baidu.picture(png_name2)
reponse = True
year = None
num = None
if len(text) >= 10:
re_reponse = re.search(re1, text)
if re_reponse:
year = re_reponse.group(1)
num = re_reponse.group(2)
os.remove(png_name2)
os.remove(png_name)
else:
reponse = False
if reponse:
pass
else:
t, fp = imagetoBig(png_name2)
if t:
baidu = BaiduApi()
text = baidu.picture(fp)
print(text)
shutil.rmtree(fp[:fp.rfind("\\")])
if len(text) >= 10:
re_reponse = re.search(re1, text)
if re_reponse:
year = re_reponse.group(1)
num = re_reponse.group(2)
os.remove(png_name2)
os.remove(png_name)
else:
reponse = False
else:
reponse = False
idn = "沪信衡估报字第G" + year + "-" + num[1:] + "号"
return idn, reponse
def shapes_from_stream(doc: fitz.Document, page: fitz.Page):
''' Get rectangle shapes, e.g. highlight, underline, table borders, from page source contents.
---
Args:
- doc: fitz.Document representing the pdf file
- page: fitz.Page, current page
The page source is represented as contents of stream object. For example,
```
/P<</MCID 0>> BDC
...
1 0 0 1 90.0240021 590.380005 cm
...
1 1 0 rg # or 0 g
...
285.17 500.11 193.97 13.44 re f*
...
214 320 m
249 322 l
...
EMC
```
where,
- `cm` specify a coordinate system transformation, here (0,0) translates to (90.0240021 590.380005)
- `q`/`Q` save/restores graphic status
- `rg` / `g` specify color mode: rgb / grey
- `re`, `f` or `f*`: fill rectangle path with pre-defined color
- `m` (move to) and `l` (line to) defines a path
In this case,
- a rectangle with:
- fill color is yellow (1,1,0)
- lower left corner: (285.17 500.11)
- width: 193.97
- height: 13.44
- a line from (214, 320) to (249, 322)
Read more:
- https://github.com/pymupdf/PyMuPDF/issues/263
- https://github.com/pymupdf/PyMuPDF/issues/225
- https://www.adobe.com/content/dam/acom/en/devnet/pdf/pdf_reference_archive/pdf_reference_1-7.pdf
'''
# Each object in PDF has a cross-reference number (xref):
# - to get its source contents: `doc.xrefObject()` or low level API `doc._getXrefString()`; but for stream objects, only the non-stream part is returned
# - to get the stream data: `doc.xrefStream(xref)` or low level API `doc._getXrefStream(xref)`
# - the xref for a page object itself: `page.xref`
# - all stream xref contained in one page: `page.getContents()`
# - combine all stream object contents together: `page.readContents()` with PyMuPDF>=1.17.0
#
# Clean contents first:
# syntactically correct, standardize and pretty print the contents stream
page.cleanContents()
xref_stream = page.readContents().decode(encoding="ISO-8859-1")
# transformation matrix for coordinate system conversion from pdf to fitz
matrix = page.transformationMatrix
# Graphic States: working CS is coincident with the absolute origin (0, 0)
# Refer to PDF reference v1.7 4.2.3 Transformation Metrices
# | a b 0 |
# [a, b, c, d, e, f] => | c b 0 |
# | e f 1 |
ACS = fitz.Matrix(0.0) # identity matrix
WCS = fitz.Matrix(0.0)
# Graphics color:
# - color space: PDF Reference Section 4.5 Color Spaces
# NOTE: it should have to calculate color value under arbitrary color space, but it's really hard work for now.
# So, consider device color space only like DeviceGray, DeviceRGB, DeviceCMYK, and set black for all others.
device_space = True
color_spaces = _check_device_cs(doc, page)
# - stroking color
Acs = utils.RGB_value((0.0, 0.0, 0.0)) # stored value
Wcs = Acs # working value
# - filling color
Acf = utils.RGB_value((0.0, 0.0, 0.0))
Wcf = Acf
# Stroke width
Ad = 0.0
Wd = 0.0
# In addition to lines, rectangles are also processed with border path
paths = [] # a list of path, each path is a list of points
# Check line by line
# Cleaned by `page.cleanContents()`, operator and operand are aligned in a same line;
# otherwise, have to check stream contents word by word (line always changes)
lines = xref_stream.splitlines()
strokes, fills = [], []
for line in lines:
words = line.split()
if not words: continue
op = words[-1] # operator always at the end after page.cleanContents()
# -----------------------------------------------
# Color Operators: PDF Reference Table 4.24
# -----------------------------------------------
# - set color space:
# color_space_name cs # specify color space
# c1 c2 ... SC/SCN # components under defined color space
if op.upper() == 'CS':
Wcs = utils.RGB_value((0.0, 0.0, 0.0))
Wcf = utils.RGB_value((0.0, 0.0, 0.0))
# Consider normal device cs only
device_space = color_spaces.get(words[0], False)
# - set color: color components under specified color space
elif op.upper() == 'SC': # c1 c2 ... cn SC
c = _RGB_from_color_components(words[0:-1], device_space)
# nonstroking color
if op == 'sc':
Wcf = c
# stroking color
else:
Wcs = c
# - set color: color components under specified color space
elif op.upper() == 'SCN': # c1 c2 ... cn [name] SC
if utils.is_number(words[-2]):
c = _RGB_from_color_components(words[0:-1], device_space)
else:
c = _RGB_from_color_components(words[0:-2], device_space)
# nonstroking color
if op == 'scn':
Wcf = c
# stroking color
else:
Wcs = c
# - DeviceGray space, equal to:
# /DeviceGray cs
# c sc
elif op.upper() == 'G': # 0 g
g = float(words[0])
# nonstroking color, i.e. filling color here
if op == 'g':
Wcf = utils.RGB_value((g, g, g))
# stroking color
else:
Wcs = utils.RGB_value((g, g, g))
# - DeviceRGB space
elif op.upper() == 'RG': # 1 1 0 rg
r, g, b = map(float, words[0:-1])
# nonstroking color
if op == 'rg':
Wcf = utils.RGB_value((r, g, b))
# stroking color
else:
Wcs = utils.RGB_value((r, g, b))
# - DeviceCMYK space
elif op.upper() == 'K': # c m y k K
c, m, y, k = map(float, words[0:-1])
# nonstroking color
if op == 'k':
Wcf = utils.CMYK_to_RGB(c, m, y, k, cmyk_scale=1.0)
# stroking color
else:
Wcs = utils.CMYK_to_RGB(c, m, y, k, cmyk_scale=1.0)
# -----------------------------------------------
# Graphics State Operators: PDF References Table 4.7
# -----------------------------------------------
# CS transformation: a b c d e f cm, e.g.
# 0.05 0 0 -0.05 0 792 cm
# refer to PDF Reference 4.2.2 Common Transformations for detail
elif op == 'cm':
# update working CS
components = list(map(float, words[0:-1]))
Mt = fitz.Matrix(*components)
WCS = Mt * WCS # M' = Mt x M
# stroke width
elif op == 'w': # 0.5 w
Wd = float(words[0])
# save or restore graphics state:
# only consider transformation and color here
elif op == 'q': # save
ACS = fitz.Matrix(WCS) # copy as new matrix
Acf = Wcf
Acs = Wcs
Ad = Wd
elif op == 'Q': # restore
WCS = fitz.Matrix(ACS) # copy as new matrix
Wcf = Acf
Wcs = Acs
Wd = Ad
# -----------------------------------------------
# Path Construction Operators: PDF References Table 4.9
# -----------------------------------------------
# rectangle block:
# x y w h re is equivalent to
# x y m
# x+w y l
# x+w y+h l
# x y+h l
# h # close the path
elif op == 're':
# ATTENTION:
# top/bottom, left/right is relative to the positive direction of CS,
# while a reverse direction may be performed, so be careful when calculating
# the corner points.
# Coordinates in the transformed PDF CS:
# y1 +----------+
# | | h
# y0 +----w-----+
# x0 x1
#
# (x, y, w, h) before this line
x0, y0, w, h = map(float, words[0:-1])
path = []
path.append((x0, y0))
path.append((x0 + w, y0))
path.append((x0 + w, y0 + h))
path.append((x0, y0 + h))
path.append((x0, y0))
paths.append(path)
# path: m -> move to point to start a path
elif op == 'm': # x y m
x0, y0 = map(float, words[0:-1])
paths.append([(x0, y0)])
# path: l -> straight line to point
elif op == 'l': # x y l
x0, y0 = map(float, words[0:-1])
paths[-1].append((x0, y0))
# close the path
elif op == 'h':
for path in paths:
_close_path(path)
# -----------------------------------------------
# Path-painting Operatores: PDF Reference Table 4.10
# -----------------------------------------------
# close and stroke the path
elif op.upper() == 'S':
# close
if op == 's':
for path in paths:
_close_path(path)
# stroke path
for path in paths:
res = _stroke_path(path, WCS, Wcs, Wd, matrix)
strokes.extend(res)
# reset path
paths = []
# fill the path
elif line in ('f', 'F', 'f*'):
for path in paths:
# close the path implicitly
_close_path(path)
# fill path
res = _fill_rect_path(path, WCS, Wcf, matrix)
fills.append(res)
# reset path
paths = []
# close, fill and stroke the path
elif op.upper() in ('B', 'B*'):
for path in paths:
# close path
_close_path(path)
# fill path
res = _fill_rect_path(path, WCS, Wcf, matrix)
fills.append(res)
# stroke path
res = _stroke_path(path, WCS, Wcs, Wd, matrix)
strokes.extend(res)
# reset path
paths = []
# TODO: clip the path
elif line in ('W', 'W*'):
pass
# end the path without stroking or filling
elif op == 'n':
paths = []
return strokes, fills
def export_data_pdf(sender, instance, created, **kwargs):
uploaded_pdf_url = instance.uploaded_pdf.path
# address_string = camelot.read_pdf(uploaded_pdf_url, flavor='stream', row_tol=9, table_areas=['50,720,780,680'])
address_string = camelot.read_pdf(uploaded_pdf_url, flavor='stream', row_tol=9, table_areas=['50,720,400,680'])
csv_address_f = os.path.join(settings.MEDIA_ROOT, 'temp', 'csv_address.csv')
csv = address_string[0].to_csv(csv_address_f)
if csv_address_f:
with open(csv_address_f, 'r', encoding='utf-8') as f:
row_read = CSV.reader(f)
for row in row_read:
pp = (row[0].strip(" '")).split(":")
if 'Квартира' in pp[0]:
hh = [int(s) for s in pp[0].split() if s.isdigit()]
global_appartment = hh[0]
print(global_appartment)
else:
print('pp')
print(pp)
print(pp[1])
adress_item_list = pp[1].split(",")
print('adress_item_list')
print(adress_item_list)
i = 0
total_list = []
for item in adress_item_list:
k = ''
v = ''
total_val = adress_item_list[i].split()
for word in total_val:
if word[0].isupper() or word[0].isdigit():
v = word
else:
k = word
if i == 1:
v = adress_item_list[i].strip()
total_list.append([k, v])
i+=1
print('total_list')
print(total_list)
total_dict = {
'city_type':'',
'city_name': '',
'street_type':'',
'street':'',
'micro_rayon':'',
'house_number':'',
'corpus_number':'',
'litera':''
}
for item in total_list:
if item[0] == 'город' or item[0] == 'поселение' or item[0] == 'деревня' or item[0] == 'поселок':
k = 'city_type'
v = item[0]
total_dict.update({k: v})
k = 'city_name'
v = item[1]
total_dict.update({k: v})
elif item[0] == 'улица' or item[0] == 'ул.' or item[0] == 'переулок' or item[0] == 'пер.' or item[0] == 'проспект' or item[0] == 'просп.' or item[0] == 'проезд' or item[0] == 'шоссе' or item[0] == 'площадь' or item[0] == 'наб.' or item[0] == 'набережная' or item[0] == 'бульвар' or item[0] == 'бул.':
k = 'street_type'
v = item[0]
total_dict.update({k: v})
k = 'street'
v = item[1]
total_dict.update({k: v})
elif item[0] == 'микрорайон':
k = 'micro_rayon'
v = item[1]
total_dict.update({k: v})
elif item[0] == 'дом':
k = 'house_number'
v = item[1]
total_dict.update({k: v})
elif item[0] == 'корпус':
k = 'corpus_number'
v = item[1]
total_dict.update({k: v})
elif item[0] == 'litera':
k = 'litera'
v = item[1]
total_dict.update({k: v})
elif item[0] == 'строение':
k = 'build_number'
v = item[1]
total_dict.update({k: v})
print('total_dict')
print(total_dict)
# func(adress_item_list[i])
v, created = Adress.objects.update_or_create(
order=instance,
defaults=total_dict,
)
if instance.new_source:
tables = camelot.read_pdf(uploaded_pdf_url)
else:
tables = camelot.read_pdf(uploaded_pdf_url, flavor='stream', row_tol=9, table_areas=['50,680,780,100'])
print(tables[0])
print(tables[0].parsing_report)
print(tables[0].df)
json_table = os.path.join(settings.MEDIA_ROOT, 'temp', 'json_table.json')
json_table2 = os.path.join(settings.MEDIA_ROOT, 'temp', 'json_table2.csv')
json = tables[0].to_json(path=json_table)
# json1 = tables[0].to_csv(path=json_table2, orient = 'records', lines = 'True')
if json_table:
with open(json_table, 'r') as f:
print("------------data-------------------")
data = JSON.load(f)
ExplicationListItem.objects.filter(order_list=instance).delete()
i = 0
data.pop()
for x in data:
if i > 2 and x['9'] != '':
ExplicationListItem.objects.create(
order_list = instance,
floor_number = x['0'],
appart_number_item = x['1'],
appart_name_item = x['2'],
square_total_item = x['3'],
square_general_item = x['4'],
square_advanced_item = x['5'],
square_logdi_item = x['6'],
square_balkon_item = x['7'],
square_another_item = x['8'],
height_item = x['9'],
apart_number = global_appartment
)
i += 1
explication_list_items = ExplicationListItem.objects.filter(order_list=instance)
def string_to_correct_decimal(string):
print('string')
print(string)
result = Decimal(string.strip(" '").replace(',', '.'))
return result
square_total_sum = Decimal("0.0")
square_general_sum = Decimal("0.0")
square_advanced_sum = Decimal("0.0")
square_logdi_sum = Decimal("0.0")
square_balkon_sum = Decimal("0.0")
square_another_sum = Decimal("0.0")
square_total_sum_global = Decimal("0.0")
for items in explication_list_items:
if items.square_total_item:
square_total_sum += string_to_correct_decimal(items.square_total_item)
# print(square_total_sum)
if items.square_general_item:
square_general_sum += string_to_correct_decimal(items.square_general_item)
# print(square_general_sum)
if items.square_advanced_item:
square_advanced_sum += string_to_correct_decimal(items.square_advanced_item)
# print(square_advanced_sum)
if items.square_logdi_item:
square_logdi_sum += string_to_correct_decimal(items.square_logdi_item)
# print(square_logdi_sum)
if items.square_balkon_item:
square_balkon_sum += string_to_correct_decimal(items.square_balkon_item)
# print(square_balkon_sum)
if items.square_another_item:
square_another_sum += string_to_correct_decimal(items.square_another_item)
# print(square_another_sum)
v, created = ExplicationSquareTotal.objects.update_or_create(
order=instance,
defaults={
'square_total_summa':square_total_sum,
'square_general_summa':square_general_sum,
'square_advanced_summa':square_advanced_sum,
'square_logdi_summa':square_logdi_sum,
'square_balkon_summa':square_balkon_sum,
'square_another_summa':square_another_sum,
'square_total_summa_global': square_total_sum + square_logdi_sum + square_balkon_sum + square_another_sum
},
)
path_img_name = 'schema_' + str(instance.order_number) + '.png'
path_img_scheme = os.path.join(settings.MEDIA_ROOT, 'uploaded_pdf/schemes/', path_img_name)
path_img_scheme_bd = "uploaded_pdf/schemes/%s" % path_img_name
current_site = Site.objects.get_current().domain
path_full_pdf = "https://%s%s" % (current_site, reverse_lazy('pdftrans:order_full_pdf_view_n', kwargs={'pk': instance.pk}))
doc = fitz.open(uploaded_pdf_url)
i = 0
for page in doc: # iterate through the pages
if i == 1:
zoom = 2 # zoom factor
mat = fitz.Matrix(zoom, zoom)
pix = page.getPixmap(matrix = mat, alpha = False) # render page to an image
pix.writePNG(path_img_scheme) # store image as a PNG
def trim(im):
bg = Image.new(im.mode, im.size, im.getpixel((0,0)))
diff = ImageChops.difference(im, bg)
diff = ImageChops.add(diff, diff, 2.0, -100)
bbox = diff.getbbox()
if bbox:
return im.crop(bbox)
im = Image.open(path_img_scheme)
im = trim(im)
im.save(path_img_scheme)
order_img_clear = OrderImage.objects.filter(order_fk=instance).delete()
v, created = OrderImage.objects.update_or_create(
order_fk=instance,
defaults={'order_image': path_img_scheme_bd, 'fullpdf_url_staff': path_full_pdf }
)
i+=1
# sending email method -=send_mail=-
path_full_pdf_for_email = str(path_full_pdf)
path_full_link_site = 'https://' + str(current_site) + '/get-order-info/' + str(instance.pk)
context = {
'order_number': instance.order_number,
'link_doc': path_full_pdf,
'link_site': path_full_link_site,
}
str_for_traslit = unidecode(str(instance.adress))
subject = str_for_traslit + ' - Док №: ' + str(instance.order_number)
from_email = '*****@*****.**'
to = '*****@*****.**'
html_content = render_to_string('mail_templates/mail_template_btiorder.html', context)
text_content = strip_tags(html_content)
msg = EmailMultiAlternatives(subject, text_content, from_email, [to])
msg.attach_alternative(html_content, "text/html")
if instance.is_emailed == False:
if subject and html_content and from_email:
try:
if msg.send():
Order.objects.filter(pk=instance.pk).update(is_emailed=True)
instance.is_emailed = True
except BadHeaderError:
return print('Invalid header found in email %s' % instance.pk)
return print('email is sended %s' % instance.pk)
else:
return print('Make sure all fields are entered and valid %s' % instance.pk)
pass
import fitz
pdf = fitz.open('./image/1.pdf')
for pg in range(0, 1):
page = pdf[pg]
rotate = int(0)
zoom_x = 2.0
zoom_y = 2.0
trans = fitz.Matrix(zoom_x, zoom_y).preRotate(rotate)
pm = page.getPixmap(matrix=trans, alpha=False)
pm.writePNG('./image/%s.png' % 1)
annot = page.addFreetextAnnot(r, t1, rotate = 90)
annot.setBorder(border)
annot.update(fontsize = 10, border_color=red, fill_color=gold, text_color=blue)
print_descr(annot.rect, annot)
r = annot.rect + displ
print("added 'FreeText'")
annot = page.addTextAnnot(r.tl, t1)
print_descr(annot.rect, annot)
print("added 'Sticky Note'")
pos = annot.rect.tl + displ.tl
# first insert 4 rotated text lines
page.insertText(pos, text, fontsize=11, morph = (pos, fitz.Matrix(-15)))
# now search text to get the quads
rl = page.searchFor("text in line", quads = True)
r0 = rl[0]
r1 = rl[1]
r2 = rl[2]
r3 = rl[3]
annot = page.addHighlightAnnot(r0)
# need to convert quad to rect for descriptive text ...
print_descr(r0.rect, annot)
print("added 'HighLight'")
annot = page.addStrikeoutAnnot(r1)
print_descr(r1.rect, annot)
print("added 'StrikeOut'")
def __init__(self, parent, filename):
defPos = wx.DefaultPosition
defSiz = wx.DefaultSize
zoom = 1.2 # zoom factor of display
wx.Dialog.__init__(self,
parent,
id=wx.ID_ANY,
title=u"Display with PyMuPDF: ",
pos=defPos,
size=defSiz,
style=wx.CAPTION | wx.CLOSE_BOX
| wx.DEFAULT_DIALOG_STYLE)
#======================================================================
# display an icon top left of dialog, append filename to title
#======================================================================
if do_icon:
self.SetIcon(ico_pdf.img.GetIcon()) # set a screen icon
self.SetTitle(self.Title + filename)
self.SetBackgroundColour(wx.Colour(240, 230, 140))
#======================================================================
# open the document with MuPDF when dialog gets created
#======================================================================
self.doc = fitz.open(filename) # create Document object
if self.doc.needsPass: # check password protection
self.decrypt_doc()
if self.doc.isEncrypted: # quit if we cannot decrpt
self.Destroy()
return
self.dl_array = [0] * len(self.doc)
self.last_page = -1 # memorize last page displayed
self.link_rects = [] # store link rectangles here
self.link_texts = [] # store link texts here
self.current_idx = -1 # store entry of found rectangle
self.current_lnks = [] # store entry of found rectangle
#======================================================================
# define zooming matrix for displaying PDF page images
# we increase images by 20%, so take 1.2 as scale factors
#======================================================================
self.matrix = fitz.Matrix(zoom, zoom) # will use a constant zoom
'''
=======================================================================
Overall Dialog Structure:
-------------------------
szr10 (main sizer for the whole dialog - vertical orientation)
+-> szr20 (sizer for buttons etc. - horizontal orientation)
+-> button forward
+-> button backward
+-> field for page number to jump to
+-> field displaying total pages
+-> PDF image area
=======================================================================
'''
# forward button
self.ButtonNext = wx.Button(self, wx.ID_ANY, u"forw", defPos, defSiz,
wx.BU_EXACTFIT)
# backward button
self.ButtonPrevious = wx.Button(self, wx.ID_ANY, u"back", defPos,
defSiz, wx.BU_EXACTFIT)
#======================================================================
# text field for entering a target page. wx.TE_PROCESS_ENTER is
# required to get data entry fired as events.
#======================================================================
self.TextToPage = wx.TextCtrl(self, wx.ID_ANY, u"1", defPos,
wx.Size(40, -1),
wx.TE_RIGHT | wx.TE_PROCESS_ENTER)
# displays total pages and page paper format
self.statPageMax = wx.StaticText(
self, wx.ID_ANY, "of " + str(len(self.doc)) + " pages.", defPos,
defSiz, 0)
self.links = wx.CheckBox(self, wx.ID_ANY, u"show links", defPos,
defSiz, wx.ALIGN_LEFT)
self.links.Value = True
self.paperform = wx.StaticText(self, wx.ID_ANY, "", defPos, defSiz, 0)
# define the area for page images and load page 1 for primary display
self.PDFimage = wx.StaticBitmap(self,
wx.ID_ANY,
self.pdf_show(1),
defPos,
defSiz,
style=0)
#======================================================================
# the main sizer of the dialog
#======================================================================
self.szr10 = wx.BoxSizer(wx.VERTICAL)
szr20 = wx.BoxSizer(wx.HORIZONTAL)
szr20.Add(self.ButtonNext, 0, wx.ALL, 5)
szr20.Add(self.ButtonPrevious, 0, wx.ALL, 5)
szr20.Add(self.TextToPage, 0, wx.ALL, 5)
szr20.Add(self.statPageMax, 0, wx.ALIGN_CENTER_VERTICAL | wx.ALL, 5)
szr20.Add(self.links, 0, wx.ALIGN_CENTER_VERTICAL | wx.ALL, 5)
szr20.Add(self.paperform, 0, wx.ALIGN_CENTER_VERTICAL | wx.ALL, 5)
# sizer ready, represents top dialog line
self.szr10.Add(szr20, 0, wx.EXPAND, 5)
self.szr10.Add(self.PDFimage, 0, wx.ALL, 5)
# main sizer now ready - request final size & layout adjustments
self.szr10.Fit(self)
self.SetSizer(self.szr10)
self.Layout()
# center dialog on screen
self.Centre(wx.BOTH)
# Bind buttons and fields to event handlers
self.ButtonNext.Bind(wx.EVT_BUTTON, self.NextPage)
self.ButtonPrevious.Bind(wx.EVT_BUTTON, self.PreviousPage)
self.TextToPage.Bind(wx.EVT_TEXT_ENTER, self.GotoPage)
self.PDFimage.Bind(wx.EVT_MOUSEWHEEL, self.OnMouseWheel)
self.PDFimage.Bind(wx.EVT_MOTION, self.move_mouse)
self.PDFimage.Bind(wx.EVT_LEFT_DOWN, self.OnLeftDown)
#------------------------------------------------------------------------------
# Main program
#------------------------------------------------------------------------------
if __name__ == "__main__":
green = getColor("limegreen")
red = getColor("red2")
doc = fitz.open()
p = doc.newPage()
img = p.newShape()
r = fitz.Rect(100, 100, 200, 200)
heart(img, r, red)
r1 = r + (100, 0, 100, 0)
p = r1.tl + (r1.br - r1.tl) * 0.5
clover(img, r1, green, morph=(p, fitz.Matrix(45)))
r2 = r1 + (100, 0, 100, 0)
diamond(img, r2, red)
r3 = r2 + (100, 0, 100, 0)
p = r3.tl + (r3.br - r3.tl) * 0.5
caro(img, r3, red, morph=(p, fitz.Matrix(45)))
r4 = r + (0, 150, 0, 150)
p = r4.tl + (r4.br - r4.tl) * 0.5
arrow(img, r4, red, morph=(p, fitz.Matrix(0)))
r5 = r4 + (r4.width, 0, r4.width, 0)
dontenter(img, r5, morph=None)
def save_to_excel(savefile, parsed_results, show_boundries, filename,
bbox_all):
tables = []
userfolder = os.path.split(os.path.dirname(savefile))[0]
if show_boundries == 'yes':
doc = fitz.open(filename)
with pd.ExcelWriter(savefile) as writer:
for page_n in parsed_results:
page_n = int(page_n)
imgname = os.path.join(userfolder, 'scanned',
'p%d.jpg' % page_n)
if not os.path.exists(imgname):
page = doc.loadPage(page_n) #number of page
mat = fitz.Matrix(1, 1)
pix = page.getPixmap(matrix=mat, alpha=False)
pix.writePNG(imgname)
if parsed_results[page_n] != []:
for idx in range(len(parsed_results[page_n])):
table_id = 'p' + str(page_n + 1) + '_id_' + str(idx +
1)
t0 = datetime.now()
imgheight, imgwidth = pdf_boundry_img(
imgname, table_id, bbox_all[page_n][idx])
print('get pdf boundry img for %s in' % table_id,
datetime.now() - t0)
table = parsed_results[page_n][idx]
table = table.applymap(lambda x: x.encode(
'unicode_escape').decode('utf-8')
if isinstance(x, str) else x)
table.to_excel(writer, sheet_name=table_id)
table.columns = table.columns.astype(str)
tables.append({
'table_id':
table_id,
'table_html':
table.to_html(index=False, index_names=False),
'imgheight':
imgheight,
'imgwidth':
imgwidth,
'remove':
False
})
doc.close()
elif show_boundries == 'ocr':
with pd.ExcelWriter(savefile) as writer:
for page_n in parsed_results:
if parsed_results[page_n] != []:
for idx in range(len(parsed_results[page_n])):
table = parsed_results[page_n][idx]
if table is None:
continue
table = table.applymap(lambda x: x.encode(
'unicode_escape').decode('utf-8')
if isinstance(x, str) else x)
table_id = 'p' + str(page_n + 1) + '_id_' + str(idx +
1)
img = plt.imread(
os.path.join(userfolder, 'scanned',
'p%d.jpg' % page_n))
imgheight, imgwidth = img.shape[0], img.shape[1]
copyfile(
os.path.join(userfolder, 'result_scanned',
'%s.jpg' % table_id),
os.path.join(userfolder, table_id + '.jpg'))
table.to_excel(writer, sheet_name=table_id)
table.columns = table.columns.astype(str)
tables.append({
'table_id':
table_id,
'table_html':
table.to_html(index=False, index_names=False),
'imgheight':
imgheight,
'imgwidth':
imgwidth,
'remove':
False
})
return tables
def get_qr_data(filename):
# global variables for qr code extraction
return_data = []
real_values = []
found_data = []
#check_values = [0, 6, 8, 11, 15, 18, 22, 24, 30, 32, 34, 36, 38, 40, 42, 45, 50]
print("Reading given %s pdf file" % filename)
doc = fitz.open(filename)
zoom = 4 # to increase the resolution
mat = fitz.Matrix(zoom, zoom)
noOfPages = doc.pageCount
print("Given pdf has %d pages" % noOfPages)
print("Starting page by page qr code detection and extraction")
for pageNo in range(noOfPages):
print("--------------Page Number %d------------------" % pageNo)
page = doc.loadPage(pageNo) # number of page
pix = page.getPixmap(matrix=mat)
output = str(
pageNo) + '.jpg' # you could change image format accordingly
print("Saving image for current page")
pix.writePNG(output)
print("Reading saved image for current page")
img = cv2.imread(output)
found = False
print("Detecting QR code...")
for delta_max_x in range(0, 200, 20):
if found:
found = False
break
for delta_max_y in range(0, 200, 20):
if found:
break
for delta in range(0, 50, 1):
y = 205 + delta
x = 2225
h = 115 - delta
w = 90
max_x = 900 - delta_max_x
max_y = 700 + delta_max_y
resized_cropped = img[y:y + h, x:x + w]
resized_cropped = cv2.resize(resized_cropped,
(max_x, max_y))
sharpen_filter = np.array([[-1, -1, -1], [-1, 10, -1],
[-1, -1, -1]])
resized_cropped = cv2.filter2D(resized_cropped, -1,
sharpen_filter)
# resized_cropped = cv2.cvtColor(resized_cropped, cv2.COLOR_BGR2GRAY)
# _, resized_cropped = cv2.threshold(resized_cropped, 10, 255, cv2.THRESH_OTSU)
qr_file = "QR_Code%d.png" % pageNo
cv2.imwrite(qr_file, resized_cropped)
hash = imagehash.average_hash(Image.open(qr_file))
otherhash = imagehash.average_hash(
Image.open('QR_Code_example.png'))
if (hash - otherhash) < 20:
barcodes = decode(resized_cropped,
symbols=[ZBarSymbol.QRCODE])
if len(barcodes) > 0:
print("Qr code detected and extracted!")
print(" %d qr code is" % pageNo, barcodes[0].data)
real_values.append(pageNo)
return_data.append([pageNo, barcodes[0].data])
found = True
found_data.append([
pageNo, y, h, delta_max_x, delta_max_y, delta
])
break
else:
if delta == 0:
found = True
break
try:
os.remove(qr_file)
os.remove(output)
except:
pass
#print("diff = ", set(check_values) - set(real_values))
pprint(found_data)
return return_data