Challenge-Response Emotion Authentication Algorithm Using Modified Horizontal Deep Learning

Abstract

Face authentication is an important biometric authentication method commonly used in security applications. It is vulnerable to different types of attacks that use authorized users' facial images and videos captured from social media to perform spoofing attacks and dynamic movements for penetrating security applications. This paper presents an innovative challenge-response emotions authentication model based on the horizontal ensemble technique. The proposed model provides high accurate face authentication process by challenging the authorized user using a random sequence of emotions to provide a specific response for every authentication trial with a different sequence of emotions. The proposed model is applied to the KDEF dataset using 10-fold cross-validations. Several improvements are made to the proposed model. First, the VGG16 model is applied to the seven common emotions. Second, the system usability is enhanced by analyzing and selecting only the four common and easy-to-use emotions. Third, the horizontal ensemble technique is applied to enhance the emotion recognition accuracy and minimize the error during authentication processes. Finally, the Horizontal Ensemble Best N-Losses (HEBNL) is applied using challenge-response emotion to improve the authentication efficiency and minimize the computational power. The successive improvements implemented on the proposed model led to an improvement in the accuracy from 92.1% to 99.27%.

A Novel Classification Model of Date Fruit Dataset Using Deep Transfer Learning

“…In this stage, the most common deep learning models [33] are scored on the dates dataset to select the best model. Deep learning methods are used to improve the accuracy of dates' classification [35,36]. The transfer learning (TL) technique that is intensively trained on the ImageNet dataset is used to benefit from the already trained model.…”

Multi-challenge database for active liveness

“…[14] locates facial landmarks in frames where the user is close to and far from the recording device, calculates the distances between these points in each frame, and uses these distances as input for a classifier. [12], [13] prompts the user to display a random sequence of emotions and classifies the images of each emotion as real or spoof using a CNN. Additionally, there are approaches based on head movements to verify facial three-dimensionality through projective invariants [53].…”

“…Another approach involves injecting information during capture and analyzing its response, such as sound or light pattern, to determine whether it interacted with a real face or a PAI [8]- [11]. A more intrusive approach involves requesting the user to solve a challenge, such as smiling, nodding, or blinking voluntarily [12], [13]. Such methods depend on the successful completion of the proposed challenges, while also leveraging their dynamic aspects to extract additional information.…”