Android with root Git for version control Lircd with Raspberry Pi for IR receiver and sender Tips for Windows Depolying your own password management tool -- KeeWeb Depoly your flask app into Heroku Fix shit IE code manually ISBN to Book Category by Scraping DangDang A Generic Makefile for C/C++ Program Configure Raspberry pi Remove watermark with PyPDF2 tips for docker Anaconda+TensorFlow+CUDA Snippets Configure Remote Mathematica Kernel Build your own ngrok server Access Array SSL VPN 使用Rstudio制作html5幻灯片 tips for Mac OS X system Tips for ipython notebook 配置Ubuntu server + Openbox (Obuntu) tips for Vimperator tips for Vim 安装CUDA My First Jekyll Blog rsync常见选项 在Linux中读取Ipod touch的文件 tip for texmacs 在VPS上建站的一些tip Gnuplot绘图札记 Samba系统和autofs自动挂载 Linux中alsamixer声卡无法录音 搭建自己的RSS订阅器——Tiny Tiny RSS Grub2引导安装Ubuntu awk tips 将Ubuntu系统装入U盘 The Great Rtorrent 编译GCC 再这样剁手!!!该死的libgd 使用ulimit进行资源限制 使用SSH代理上IPV6 使用RCurl抓取网页数据 修复Ubuntu Grub记 openbox中的文件关联 在Ubuntu 12.04下编译qtiplot 处理BCM4312网卡驱动纪实 配置我的Ubuntu Server记 Cygwin杂记 Linux 使普通用户具有以超级权限执行脚本 让firefox自定义地处理文件类型 WordPress优秀主题及插件 在phpcloud上搭建wordpress UBUNTU下用pptpd做VPN server ubuntu升级内核过后的一些问题 安装telnet服务 kubuntu札记 64位kubuntu札记 统计软件R Virtualbox stardict星际译王 Ubuntu重装windows系统后的grub引导修复 SSH服务及花生壳域名解析 采用cbp2make工具由code::blocks工程创建makefile文件 UBUNTU 札记

Remove watermark with PyPDF2

2018年06月19日

You may find the watermark of PDF file very annoying and it usually very hard to remove. With the power of python package PyPDF2 it would be done with some research.

PyPDF2 can analysis the source PDF file and identify all elements, in which some may correspond to the watermark. To be noticed, the watermark can be composed of several objects.

Like following code, you can analysis the structure of PDF file to identify the watermark objects.

from PyPDF2 import PdfFileReader, PdfFileWriter
from PyPDF2.pdf import ContentStream
from PyPDF2.generic import TextStringObject, NameObject
from PyPDF2.utils import b_
import numpy as np

file = open('/tmp/深度学习+中文版.pdf', "rb")

output = PdfFileWriter()
source = PdfFileReader(file)

print(source.getDocumentInfo())


def guess_codes(s):
    avaiable_c = []

    codes = 'ascii,big5,big5hkscs,cp037,cp273,cp424,cp437,cp500,cp720,cp737,cp775,cp850,cp852,cp855,cp856,cp857,cp858,cp860,cp861,cp862,cp863,cp864,cp865,cp866,cp869,cp874,cp875,cp932,cp949,cp950,cp1006,cp1026,cp1125,cp1140,cp1250,cp1251,cp1252,cp1253,cp1254,cp1255,cp1256,cp1257,cp1258,cp65001,euc_jp,euc_jis_2004,euc_jisx0213,euc_kr,gb2312,gbk,gb18030,hz,iso2022_jp,iso2022_jp_1,iso2022_jp_2,iso2022_jp_2004,iso2022_jp_3,iso2022_jp_ext,iso2022_kr,latin_1,iso8859_2,iso8859_3,iso8859_4,iso8859_5,iso8859_6,iso8859_7,iso8859_8,iso8859_9,iso8859_10,iso8859_11,iso8859_13,iso8859_14,iso8859_15,iso8859_16,johab,koi8_r,koi8_t,koi8_u,kz1048,mac_cyrillic,mac_greek,mac_iceland,mac_latin2,mac_roman,mac_turkish,ptcp154,shift_jis,shift_jis_2004,shift_jisx0213,utf_32,utf_32_be,utf_32_le,utf_16,utf_16_be,utf_16_le,utf_7,utf_8,utf_8_sig'.split(
        ',')
    if not isinstance(s, bytes):
        for c in codes:
            try:
                s.decode(c)
                print(c)
                avaiable_c.append(c)
            except Exception:
                pass

    return avaiable_c



target_locations = np.array([
    [0.70711, 0.70711, -0.70711, 0.70711, -308.753, -165.279],
    [1, 0, 0, 1, -44.831, 54.605]])


def match_location(location, target, epsilon=1e-5):
    # targe must be n*6 numpy matrix
    return np.any(np.abs(np.array([i.as_numeric() for i in location]) - target).max(axis=1) < epsilon)


for p in range(source.getNumPages()):
    page = source.getPage(p)
    # print(page.extractText())
    content_object, = page["/Contents"].getObject()

    content = ContentStream(content_object, source)
    for operands, operator in content.operations:
        # print(operator, operands) # pdf元素的类型和值
	
	# 主要的代码在这里,使用各种方式找到水印可识别的特征
        # if operator == b_("TJ"): # `b_`只是python2/3中bytes类型转换的冗余代码
        #     text = operands[0][0]
        #     # if isinstance(text, bytes):
        #     #     print('====  ', text, '  ====')
        #     #     for c in guess_codes(text):
        #     #         print(c, text.decode(c))
        #     if isinstance(text, TextStringObject) and text in target_str:
        #         operands[0] = TextStringObject('')

        if operator == b_("cm") and match_location(operands, target_locations):
            operands[:] = []

    page.__setitem__(NameObject('/Contents'), content)
    output.addPage(page)


outputStream = open("/tmp/output.pdf", "wb")
output.write(outputStream)
outputStream.close()

file.close()

Some very simple watermarks may be just plain text. But some can be very complex. With the help of InkScape, one may find out the specific signeture of the watermark very easily. Import several pages into InkScape and use the Edit -> XML editor to view the structure of the pages.

Some svg node may have a transform matrix,use that to identity the PyPDF2 objects. Set the matrix of these object to empty will make the watermark disappear!

If you read pdf file into PdfFileReader and modifiy every page, then insert them into a PdfFileWriter you may loss the bookmark information, like what have been done in upper code. A better choice is to extent the PdfFileMerger class because it will automatically handle the bookmarks and other information. Some may suggest using the cloneDocumentFromReader method, but it may generate blank pages.

from PyPDF2 import PdfFileMerger
from PyPDF2.pdf import ContentStream
from PyPDF2.utils import b_
from PyPDF2.generic import NameObject
import numpy as np


class PdfFileModifyMerger(PdfFileMerger):
    def merge(self, *args, after_page_append=[], **kwargs):
        super(PdfFileModifyMerger, self).merge(*args, **kwargs)

        pdfr = self.inputs[-1][1]

        position = args[0]
        pages = kwargs.get('pages', None)
        if pages is None:
            pages = (0, pdfr.getNumPages())
        elif isinstance(pages, PageRange):
            pages = pages.indices(pdfr.getNumPages())
        elif not isinstance(pages, tuple):
            raise TypeError('"pages" must be a tuple of (start, stop[, step])')

        for mp in self.pages[position:(position+len(range(*pages)))]:
            for mf in after_page_append:
                mp.pagedata = mf(mp.pagedata, pdfr)


def cm_remover(locations, epsilon=1e-5):
    # locations must be n*6 numpy matrix
    def remove_cm(pg, pdfr):
        def matched_location(loc):
            return np.any(np.abs(np.array(loc) - locations).max(axis=1) < epsilon)
        content_object, = pg["/Contents"].getObject()

        content = ContentStream(content_object, pdfr)
        for operands, operator in content.operations:
            if operator == b_("cm") and matched_location([i.as_numeric() for i in operands]): # this is where magic happens
                operands[:] = []

        pg.__setitem__(NameObject('/Contents'), content)
        return pg

    return remove_cm


target_locations = np.array([ # find following information with InkScape, comparing the PyPDF2 outputs
    [0.70711, 0.70711, -0.70711, 0.70711, -308.753, -165.279],
    [1, 0, 0, 1, -44.831, 54.605]])


output = PdfFileModifyMerger()

output.merge(len(output.pages),
             '/tmp/深度学习+中文版.pdf',
             after_page_append=[cm_remover(target_locations)])

with open("/tmp/output.pdf", "wb") as outputStream:
    output.write(outputStream)