Deep Learning Based CAPTCHA Recognition Network with Grouping Strategy.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Websites use CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) to prevent bot attacks.
• Text-based CAPTCHAs are common but increasingly vulnerable to advanced deep learning models.
• Convolutional Neural Networks (CNNs) have shown significant progress in image recognition tasks.

Purpose of the Study:

• To develop an effective and efficient method for recognizing text-based CAPTCHAs.
• To address the challenge of deep learning models bypassing traditional CAPTCHA security.
• To present a CNN-based approach that is computationally efficient and requires minimal storage.

Main Methods:

• Proposed a novel CAPTCHA recognition method using CNNs.
• Generated duplicate CAPTCHA images with binary character location encoding.
• Fed replicated images sequentially into a trained CNN for character output.
• Designed a straightforward CNN architecture avoiding individual character segmentation.

Main Results:

• The proposed CNN model demonstrated high accuracy in recognizing CAPTCHA characters.
• The method proved effective despite advancements in deep learning capabilities.
• The model's simple architecture and low storage requirements were validated.

Conclusions:

• The presented CNN-based CAPTCHA recognition method is effective and accurate.
• This approach offers a viable solution for enhancing website security against automated attacks.
• The method provides an efficient alternative to traditional CAPTCHA segmentation techniques.