Recognition of CAPTCHAs Utilizing The Middle Part of The Convolutional Feature Map

CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart) is an essential method for distinguishing between humans and machines, utilized by websites to prevent automated malicious software assaults. Examining CAPTCHA recognition may expose vulnerabilities in CAPTCHA systems. The fundamental objective of CAPTCHAs can be undermined by utilizing deep learning and computer vision methodologies. A Deep Convolutional Neural Network model is utilized to recognize CAPTCHAs, obviating the necessity for conventional image processing methods like localization and segmentation. Our research introduces a CAPTCHA recognition system that emphasizes the core region of feature maps via a customized DCNN model integrated with an attention mechanism. This method facilitates the extraction of essential character information required for training within the intricate environment of CAPTCHAs characterized by significant noise. The experimental results demonstrate that our model possesses outstanding identification capabilities for CAPTCHAs featuring background noise and character adhesion distortion. It attains exceptional precision and a minimal character error rate across multiple datasets.