Optimal Feature Selection-Based Face Liveness Detection Using Fused Long Short-Term Memory with Gated Recurrent Unit
K. Sathish Kumar, N. Jyothi · International Journal of Image and Graphics · 2025
Face-recognizing technology has been extensively utilized in individual authentication systems for successful performance. Although many investigations have demonstrated that recognized faces may be weak in an aggressive setting, where a competitor might pose as an authorized user to trick the technology yet, the system’s safety may become a critical concern. Although many ways to discern between genuine and synthetic faces are being presented, they’re laborious, costly, or dependent on noises and brightness. In existing, liveness detection becomes a crucial task and also it prone to attack based on some kind of spoofing materials. Thus, it makes the system more complex, and the detection of face liveness needs to be considered promptly. Henceforth, the research contributes to developing efficient deep learning models to tackle the complexities present in the existing conventional models to provide a better face-liveness detection model. At first, the images related to the face are attained from the benchmark resources. Then, the acquired images are utilized to perform nonlinear diffusion in the image. Additionally, the nonlinear diffused images are offered in the feature extraction phase. Here, the features are attained by deep learning techniques like VGG16, Residual Attention Network (RAN), and MobileNet. Further, the gathered features are provided to the weighted feature selection phase and the network parameters are tuned by the Adaptive Hybrid Mud Ring with Fruit Fly Optimization Algorithm (AHMR-FOA). At the end, the attained optimal selected features are fed to the face liveness detection phase. Here, effective face liveness detection is performed by Fused Long Short-Term Memory with a Gated Recurrent Unit (FLSTM-GRU) and secured better face liveness detection rate. Here, the hybrid optimization algorithm is used to tune the parameters of both LSTM and GRU. Thus, the developed face liveness detection model is secure and improves the performance rate than the traditional models through the experiment. Overall, the performance analysis of the proposed model performs enriched performance where it shows a value of 97 in terms of accuracy, specificity, sensitivity, and precision. Also, the developed model attains a value of 95 regarding the MCC measure. This analysis of the recommended model offers better detection performance in face liveness when compared with traditional methods.