DOCR-CAPTCHA: OCR Classifier based Deep Learning Technique for CAPTCHA Recognition
Amal Mathew, Apeksha Kulkarni, Anson Antony, Shreeanant Bharadwaj, Swapnil Bhalerao · 2021
The objective of this research is to analyze the security flaws of CAPTCHA generating model in order to build more resilient CAPTCHAs without such risks associated with human attempt and fail attempts. In this work, a more efficient DOCR-CAPTCHA model has presented which is deep learning approach based on an optical character recognition (OCR) to address the concerns of lower efficiency and inadequate performance of existing CAPTCHA detection algorithms. First, the DOCR-CAPTCHA model preprocesses the images to enhance the quality by following gray scale conversion, cropping and resizing the image. Second, it extracts the character to create a dictionary and mapping each character with labeling. Next, it performs classification using OCR technique and train the model. It also performed the validation on the same data. The simulation has done on the CAPTCHA images dataset. The recognition rate of this simulated model results achieved a high accuracy rate of 99.98 percent and minimized error rate of 0.0051 for the CAPTCHA train dataset. It has also compared with the existing YOLO technique and found that it has outperformed than YOLO.