Synergizing Handcrafted and Deep Features for Enhanced Face Presentation Attack Detection

This paper presents an integrated approach to enhance face presentation attack detection (PAD) by combining handcrafted Multi-level Local Binary Patterns (MLBP) and VGG16 deep learning features. Principal Component Analysis (PCA) is employed to reduce the dimensionality of the feature space, ensuring an efficient detection process. MLBP captures intricate details and patterns crucial for distinguishing between genuine and forged facial presentations. By analyzing texture patterns at multiple scales and focusing on both local texture variations and global features, MLBP effectively identifies specific characteristics that are not easily discernible with global features alone. VGG16, a convolutional neural network with 16 layers, is renowned for its deep architecture, which extracts high-level hierarchical features from facial images. It captures complex and hierarchical representations of image data, effectively differentiating between subtle variations in facial images. The features are integrated and then processed using PCA to reduce dimensionality, enhancing overall efficiency. Extensive experimentation is done on various datasets, including CS-MAD Mobile, CASIA FASD, and NUAA, demonstrating the success of the proposed approach. We have also conducted a comparative analysis with state-of-the-art literature. The comparison results demonstrate that this integrated approach distinguishes between genuine and fraudulent facial presentations more effectively than well-known methods in the field.