How to Bypass Captchas with Selenium and Tesseract: A Step‑by‑Step Python Guide

Introduction

When crawling websites, captchas are a common obstacle. This article demonstrates how to use Selenium to handle pop‑ups and solve a simple numeric captcha on an instrument reservation platform.

The captcha consists of colored digits with a simple background noise, which can be solved without AI by applying a binary threshold and feeding the result to Google’s tesseract‑OCR engine.

Note: Configuration for selenium and tesseract is assumed to be handled by the reader.

Python Practice

import re
# image processing
from PIL import Image
# text recognition
import pytesseract
# browser automation
from selenium import webdriver
import time

Solving Pop‑up Dialogs

First open the target site:

url = 'http://lims.gzzoc.com/client'
driver = webdriver.Chrome()
driver.get(url)
time.sleep(30)

The site shows an alert‑style pop‑up. Common dialog types are:

alert(message) – simple OK alert confirm(message) – OK/Cancel dialog prompt(text, defaultText) – input dialog

Non‑traditional alerts may be embedded in div, iframe, or separate windows and require element location or frame/window switching.

Capturing the Captcha Image

Steps to capture and crop the captcha:

Locate the captcha image element. Get its size and position. Take a full‑page screenshot. Crop the screenshot to the captcha region.

img = driver.find_element_by_xpath('//img[@id="valiCode"]')
time.sleep(1)
location = img.location
size = img.size
left = 2 * location['x']
top = 2 * location['y']
right = left + 2 * size['width'] - 10
bottom = top + 2 * size['height'] - 10
driver.save_screenshot('valicode.png')
page_snap_obj = Image.open('valicode.png')
image_obj = page_snap_obj.crop((left, top, right, bottom))
image_obj.show()

Adjust the scaling factor if the screenshot is offset due to browser zoom.

Binary Threshold Processing

The threshold (e.g., 205) separates foreground digits from background noise.

img = image_obj.convert("L")  # convert to grayscale
pixdata = img.load()
w, h = img.size
threshold = 205
for y in range(h):
    for x in range(w):
        if pixdata[x, y] < threshold:
            pixdata[x, y] = 0
        else:
            pixdata[x, y] = 255

Further clean isolated noise:

data = img.getdata()
w, h = img.size
black_point = 0
for x in range(1, w-1):
    for y in range(1, h-1):
        mid_pixel = data[w*y + x]
        if mid_pixel < 50:
            top_pixel = data[w*(y-1) + x]
            left_pixel = data[w*y + (x-1)]
            down_pixel = data[w*(y+1) + x]
            right_pixel = data[w*y + (x+1)]
            if top_pixel < 10: black_point += 1
            if left_pixel < 10: black_point += 1
            if down_pixel < 10: black_point += 1
            if right_pixel < 10: black_point += 1
            if black_point < 1:
                img.putpixel((x, y), 255)
            black_point = 0
img.show()

Text Recognition

result = pytesseract.image_to_string(img)
regex = '\d+'
result = ''.join(re.findall(regex, result))
print(result)

Submitting Credentials

driver.find_element_by_name('code').send_keys(result)
driver.find_element_by_name('userName').send_keys('xxx')
driver.find_element_by_name('password').send_keys('xxx')
# click confirm button
driver.find_element_by_xpath("//div[@class='form-group login-input'][3]").click()

Because binary captcha recognition is not 100% reliable, a retry loop can be used:

while True:
    try:
        # ... perform login steps ...
        break
    except:
        driver.find_element_by_id('valiCode').click()

Conclusion

After a successful login, the session cookies can be used for further automation with Selenium or passed to requests for data extraction.