CAPTCHAs against meddler image identification based on a convolutional neural network

To strengthen personal identity security through a password, the CAPTCHA authenticates the human user. In this, text-based Hue Icon Montage (HIM) and image-based Hue Icon Montage Click on meddler image and color (HIMC) CAPTCHA are displayed on a panel with a grid of images called a photomontage that contains a set of image groups that each holds six images. Fake facial images (Meddler images) are created by compositing human and animal images called ANTHROMORPH images to create CAPTCHA images against the threat of deep learning. The main contribution to this paper consists of 1) a novel CAPTCHA created with meddler images. 2) this work apply to test whether these images are strong or identified by the machine, the latest manipulated face convolution neural network MANFA framework with Xtreme Gradient Boosting (XGB-MANFA) to overcome the imbalanced dataset problem. 3) A voluminous meddler image dataset (ANTHROMORPH) is created, collected, and validated. This model produces higher security because the success probability rate of HIM and HIMC against brute-force attacks is 3.960 × 10−21. Further, the robustness of meddler images against artificial machine attacks is verified using the XGB-MANFA network, which produces only 26% accuracy that ensures the strength against machine attacks. A user study was conducted to measure the usability of HIM and HIMC CAPTCHAs with 112 participants. The proposed model’s effectiveness, efficiency, learnability, and user satisfaction scores were 0.16, 10.70, 8.40, and 8.62, respectively. The experimental result outperforms well than the existing CAPTCHAs like KCAPTCHA, reCAPTCHA, farett-gender, farett-gender and age.