当前位置：网站首页>[font anti crawl] cat x-eye Yingshi, we're bullying you again, using OCR recognition technology

[font anti crawl] cat x-eye Yingshi, we're bullying you again, using OCR recognition technology

2022-07-21 09:29:00 【InfoQ】

You are reading
【 Dream eraser 】
The blog of Finished reading , You can praise it with a little hand Find the error , Correct it in the direct comment area The second part of the eraser <font color=red>648</font> Original blog

️ Python Anti climbing actual combat scene

This blog begins , We officially entered the battlefield of font anti climbing , Today's target site is cat's eye , A classic font example , Cases are for learning purposes only , Do not use for illegal purposes .

After entering the home page of the website , Randomly select a film and television work , Enter details page .

In the response of the page, you can see , Digital related information cannot be obtained directly .

If in reptile analysis logic , The above content appears , All belong to font anti climbing sites .

We were using the

Element

Grab the element content , You can see that the target data refers to a special font style

mtsi-font

Retrieve the text in the web source code , Get the following ：

@font-face{font-family: &quot;mtsi-font&quot;;src:url(&quot;//s3plus.meituan.net/v1/mss_73a511b8f91f43d0bdae92584ea6330b/font/e5389ac6.eot&quot;);src:url(&quot;//s3plus.meituan.net/v1/mss_73a511b8f91f43d0bdae92584ea6330b/font/e5389ac6.eot?#iefix&quot;) format(&quot;embedded-opentype&quot;),url(&quot;//s3plus.meituan.net/v1/mss_73a511b8f91f43d0bdae92584ea6330b/font/e5389ac6.woff&quot;);}
 .stonefont {
 font-family: mtsi-font;
 }

It contains 2 Font file , The suffixes are

.eot

.woff

, The subsequent breakthrough point is also here , At the same time, I found beauty xxx Group related information , Looks beautiful xxx The group's site , All have similar fonts to crawl back , You can study it later .

Preference

.woff

File download analysis , Also, the name of the file should be dynamic , The file name needs to be dynamically obtained for the preparation of subsequent crawlers .

woff Full name Web Develop font formats （Web Open Font Format）, Is a web page font format standard .

After downloading the file , have access to

FontCreator

The software opens the file .

Directly drag the font file into the software , You can check , The results are as follows .

null

Please pay attention to the correspondence between codes and numbers , Then we are in Python Resolve these relationships in , So as to obtain the text information .

Looking up the font encrypted file just now , Get what's shown in the figure below .

The above code

<span class="stonefont">.</span>

Is to remove the anti climbing position , among



Mainly after watching 4 position , By comparing the two pictures, we can know , The font code corresponding to this value is

uniE36C

, The digital 9 , So when writing the font decryption code later , You need to convert the obtained text encoding into font encoding .


- >
uniE36C

So let's use Python The module in reads the font , Try to get the font encoding content .

install

fontTools

modular , Used to parse font files .

pip install fontTools -i https://pypi.tuna.tsinghua.edu.cn/simple

Manual Download 2 Font file , Then run the following code ：

from fontTools.ttLib import TTFont

TTFont('03337c30.woff').saveXML('font1.xml')
TTFont('e5389ac6.woff').saveXML('font2.xml')

obtain 2 individual XML file , It contains the data after font parsing .

Here is the abbreviation of font number , The specific drawing content of the font is at the bottom of the code , By comparison, you can find the following .

Compare the numbers in the two files 3 The coding , You can see the difference in detail , A conclusion reached at this point is ,MAOYAN In addition to font coding changes , The coordinates of font rendering have also changed .

null

Next, complete the first step of this case , Download a reference font file , For example, the content obtained above .

Then build a dictionary , Used to create the correspondence between numbers and codes .

font_dict = {
 'uniE415': '3',
 'uniF41A': '8',
 'uniF078': '7',
 'uniE5BF': '0',
 'uniE36C': '9',
 'uniF8DF': '1',
 'uniE5A5': '6',
 'uniEF4D': '4',
 'uniE6E0': '2',
 'uniED3D': '5',
}

Write the number extraction code below .

from fontTools.ttLib import TTFont

base_font = TTFont('03337c30.woff')
print(base_font)

base_list = base_font.getGlyphOrder()[2:]

print(base_list)

#  Suppose the score string captured is as follows 
# &#xe6e0;&#xe5bf;&#xe415;&#xf078;&#xf078;
font_str = '&#xe6e0;&#xe5bf;&#xe415;&#xf078;&#xf078;'

font_list = font_str.split(';')[:-1]
font_list = ['uni' + _[3:].upper() for _ in font_list]

font_dict = {
 'uniE415': '3',
 'uniF41A': '8',
 'uniF078': '7',
 'uniE5BF': '0',
 'uniE36C': '9',
 'uniF8DF': '1',
 'uniE5A5': '6',
 'uniEF4D': '4',
 'uniE6E0': '2',
 'uniED3D': '5',
}

real_numm = [font_dict[f] for f in font_list]
print(real_numm)

The output content is

['2', '0', '3', '7', '7']

, The content can correspond to the data displayed on the web page , Indicates that the basic parsing has been completed , If there is no change in the font file encoding in the web page , At this point, we have the problem , But the cat's eye is more difficult 2 A level , It changes every time it refreshes the page , Natural coding logic is much more difficult .

The next step is to enter the identification of numbers , Because the cat's eye font file is loaded dynamically , And the coding and order will change , So we use OCR Image recognition technology .

️ Actual combat scene of digital recognition

Download one before testing

woff

file , Then convert it to

ttf

file , This file is used for subsequent rendering .

Font file format conversion

from fontTools.ttLib import TTFont
from fontTools.ttLib.woff2 import decompress

#  Font files downloaded from the target site 
woff_path = &quot;./03337c30.woff&quot;
# ttf  Format file 
ttf_path = './font.ttf'
#  take  woff Document conversion  ttf  file 
decompress(woff_path, ttf_path)

font = TTFont(ttf_path)

print(font)

After running the code , obtain ttf file .

Next use PIL The module draws a font picture .

Draw an empty picture

#  Picture width and height 
img_size = 512
#  Instantiate an image object 
img = Image.new('1', (img_size, img_size), 255)
#  Drawing pictures 
draw = ImageDraw.Draw(img)
img.show()

![【 Font reverse crawling 】 cat X eye YingShi, We're bullying you again , Yes OCR Recognition technology ](https://img-blog.csdnimg.cn/51f39570ff9040c59e3c003068a9b431.png =200x)

Analyze the corresponding relationship of encoding in font file .

font = TTFont('./font.ttf')

for cmap_code, glyph_name in font.getBestCmap().items():
 print(cmap_code,glyph_name)

After the code runs , The corresponding relationship is as follows

120 x
58220 uniE36C
58389 uniE415
58789 uniE5A5
58815 uniE5BF
59104 uniE6E0
60733 uniED3D
61261 uniEF4D
61560 uniF078
62490 uniF41A
63711 uniF8DF

At this time, I found

uni

The code at the beginning , And there's a bunch of numbers in front , Next, draw this number .

from fontTools.ttLib import TTFont
from fontTools.ttLib.woff2 import decompress
from PIL import ImageFont, Image, ImageDraw

#  Picture width and height 
img_size = 512

font = TTFont('./font.ttf')
font_img = ImageFont.truetype('./font.ttf', img_size)

for cmap_code, glyph_name in font.getBestCmap().items():
 # print(cmap_code,glyph_name)

 #  Instantiate an image object 
 img = Image.new('1', (img_size, img_size), 255)

 #  Drawing pictures 
 draw = ImageDraw.Draw(img)
 #  Read the encoding into bytes 
 txt = chr(cmap_code)

 x, y = draw.textsize(txt, font=font_img)

 draw.text(((img_size - x) // 2, (img_size - y) // 2), txt, font=font_img, fill=0)

 img.show()

Run the above code , The program will show all kinds of pictures drawing numbers in turn , Later we use

ddddocr

Recognition can be completed .

The next step is to identify the part of the code , Read the binary byte stream generated by the number into the program , Directly identify .

bytes_io = BytesIO()
img.save(bytes_io, format=&quot;PNG&quot;)
word = ocr.classification(bytes_io.getvalue()) #  Identify Fonts 
print(word)

The full code is shown below ：

from fontTools.ttLib import TTFont
from fontTools.ttLib.woff2 import decompress
from PIL import ImageFont, Image, ImageDraw
from io import BytesIO
import ddddocr
&quot;&quot;&quot;
#  Font files downloaded from the target site 
woff_path = &quot;./03337c30.woff&quot;
# ttf  Format file 
ttf_path = './font.ttf'
#  take  woff Document conversion  ttf  file 
decompress(woff_path, ttf_path)

font = TTFont(ttf_path)
&quot;&quot;&quot;

#  Picture width and height 
img_size = 512

font = TTFont('./font.ttf')
font_img = ImageFont.truetype('./font.ttf', img_size)
ocr = ddddocr.DdddOcr()
for cmap_code, glyph_name in font.getBestCmap().items():
 # print(cmap_code,glyph_name)

 #  Instantiate an image object 
 img = Image.new('1', (img_size, img_size), 255)

 #  Drawing pictures 
 draw = ImageDraw.Draw(img)
 #  Read the encoding into bytes 
 txt = chr(cmap_code)

 x, y = draw.textsize(txt, font=font_img)

 draw.text(((img_size - x) // 2, (img_size - y) // 2), txt, font=font_img, fill=0)
 bytes_io = BytesIO()
 img.save(bytes_io, format=&quot;PNG&quot;)
 word = ocr.classification(bytes_io.getvalue()) #  Identify Fonts 
 print(cmap_code, glyph_name, word)

After running , Get the following output

2 extra bytes in post.stringData array
120 x
58220 uniE36C 9
58389 uniE415 3
58789 uniE5A5 6
58815 uniE5BF 0
59104 uniE6E0 2
60733 uniED3D 5
61261 uniEF4D 4
61560 uniF078 7
62490 uniF41A 8
63711 uniF8DF 1

At this time, in combination with the above , Put two parts of code （ The first part is to



Convert to

uni code

） Integrate... Together , Then complete the cat's eye font anti crawling recognition case .

Go to the cat's eye again and get a font file randomly , The following results are successfully parsed , This anti climbing case has been successfully solved .