In the growing context of online exam surveillance to ensure academic integrity, biometric authentication through facial recognition has become a common practice. However, the efficacy of this method is being questioned due to the potential vulnerability associated with the use of printed images to bypass the monitoring system. This vulnerability raises significant concerns regarding the security and authenticity of online assessments, thereby necessitating a deeper exploration of more reliable and secure facial liveness detection methods. In this study, we proposed a real-time approach for detecting facial liveliness within an anti-fraud device during online exams, leveraging facial recognition technology. Our focus was on employing a convolutional neural network algorithm to extract distinctive facial features. Convolutional neural networks, known for their adeptness in pattern detection and recognition, were at the core of our investigation. We delved into analyzing facial liveliness through two distinct approaches. Firstly, we meticulously examined facial texture, studying a dataset comprising both genuine faces and reproductions on various media such as fabric or masks. Concurrently, we implemented a method centered on detecting eye blinking. Regarding the implementation with the neural network algorithm, the results unveiled a precision rate of 57% for skin texture analysis, highlighting the inherent challenges of this method. Conversely, the eye blinking approach exhibited significantly better performance, with a precision of 96%, emphasizing its strong potential in detecting facial liveliness.