Captcha Bypass
Oleksii Gavrylenko
Data Scraper
Data Analyst
AI Application Developer
OpenCV
Python
Upwork
Other side of the Upwork
This task I got from Upwork, where I was already contracted with Scale AI.
What was the purpose of the task?
The task was to create a web scraper for a automobile details' webiste. This website is mainly focused on company owners and isn't availble for non-corporate clients.
Unfortunatelly I did not finish the project, since the client disappeared and failed to communicate further for two months.
Technologies used
- Python
- dotenv
- super_image
- PIL
- cv2
- scipy.ndimage
- pytesseract
- numpy
- pickle
- datetime
- urllib
- random
- logging
- requests
- bs4
- json
What have I achieved?
Even though the project came to an end before the work was fully complete, I managed to create an automative login solution with an AI captcha solver.
Here is the code snippet of the program I've created:
#The hardest part, the part of solving the captcha
def request_captcha():
#Getting the timestamp of current time
now = datetime.now()
timestamp = now.strftime("%Y%m%d_%H%M%S")
#Getting the link which will get us a captcha
#We need to remember, that we need to have our Basic Cookies, which should have been requested before in order to successfully solve the captcha and log in
captcha_url = os.getenv("captcha_url")
#Creating a unique captcha URL
captcha_url = f"{captcha_url}?0.{random_num()}"
#Getting the image and saving it as captcha_image.png
response = session.get(captcha_url, stream=True, timeout=10)
image_path = os.path.join(captcha_dir, f"original_{timestamp}.png")
if os.path.exists(image_path):
os.remove(image_path)
with open(image_path, "wb") as image_file:
image_file.write(response.content)
#Proccessing the image, so it is clearly seen what characters and numbers are displayed
th1 = 140
sig = 1.3 # the blurring sigma
image = Image.open(image_path)
black_and_white = image.convert("L") # converting to black and white
bw_path = os.path.join(captcha_dir, f"black_white_{timestamp}.png")
black_and_white.save(bw_path)
first_threshold = black_and_white.point(lambda p: p > th1 and 255)
ft_path = os.path.join(captcha_dir, f"first_threshold_{timestamp}.png")
first_threshold.save(ft_path)
blur = numpy.array(first_threshold) # create an image array
blurred = gaussian_filter(blur, sigma=sig)
blurred = Image.fromarray(blurred)
blur_path = os.path.join(captcha_dir, f"blurred_{timestamp}.png")
blurred.save(blur_path)
model = EdsrModel.from_pretrained('eugenesiow/edsr-base', scale=4)
inputs = ImageLoader.load_image(Image.open(blur_path))
preds = model(inputs)
scaled_path = os.path.join(captcha_dir, f"scaled_4x_{timestamp}.png")
ImageLoader.save_image(preds, scaled_path)
sharpen_filter=numpy.array([[-1,-1,-1],
[-1,9,-1],
[-1,-1,-1]])
# applying kernels to the input image to get the sharpened image
original_image=cv2.imread(scaled_path)
sharp_image=cv2.filter2D(original_image,-1,sharpen_filter)
#Identifying characters
captcha_text = pytesseract.image_to_string(sharp_image, lang='eng',
config='--psm 7 --oem 3 -c tessedit_char_whitelist=0123456789QWERTYUIOPASDFGHJKLZXCVBNM').strip()
#Info about the process
print(f"Got the captcha: {captcha_text}")
logging.debug(f"Got the captcha: {captcha_text}")
return captcha_text
