Breaking text-based CAPTCHAs with variable word and character orientation

摘要: A novel approach for automatic segmentation and recognition of CAPTCHAs with variable orientation random collapse overlapped characters is presented in this paper. Additionally, the extension proposed to break reCAPTCHA version 2012 also discussed. The original proposal consists straightening word CAPTCHA exploiting then a three-color bar code their segmentation. straightened whole provided by SVM-based learning classifier. main goal research reduce vulnerability from spam frauds as well provide an recognizing either handwritten or degraded damaged texts ancient manuscripts OCR systems. designed framework breaking has been tested achieving average success rate up 82% 2011 95.5% extended response time less than 0.5s per two-word reCAPTCHA. implemented SVM classifier shows competitive precision about 94%. obtained very satisfactory results confirm that may be used development new security mechanisms protect users against cyber-criminal activities Internet threats. Automatic Web sites proposed.Anti-recognition techniques use collapsed orientation.Aligned are segmented code.Original provides real-time recognition.Extended beating better performance.