Time-Aware Face Anti-Spoofing with Rotation Invariant Local Binary Patterns and Deep Learning

TL;DR

This work tackles the challenge of rapid, reliable face liveness detection on commodity mobile devices by combining a rotation-invariant color texture feature, based on $LBP_{P,R}^{riu2}$, with color histogram data and time-aware deep learning. The model processes short sequences ($n=16$ frames) to form a temporal representation, then uses an LSTM-based network to classify live versus spoofed attempts, achieving high accuracy with lightweight data transfer. Extensive experiments on OULU-NPU, RECOD-MPAD, and SiW show that fusing color-space histograms with texture features yields strong performance, and cross-dataset tests indicate good generalization, especially when leveraging majority voting across sequences. The approach emphasizes practical deployment, balancing accuracy, latency, and bandwidth, making it well-suited for real-world mobile authentication systems.

Abstract

Facial recognition systems have become an integral part of the modern world. These methods accomplish the task of human identification in an automatic, fast, and non-interfering way. Past research has uncovered high vulnerability to simple imitation attacks that could lead to erroneous identification and subsequent authentication of attackers. Similar to face recognition, imitation attacks can also be detected with Machine Learning. Attack detection systems use a variety of facial features and advanced machine learning models for uncovering the presence of attacks. In this work, we assess existing work on liveness detection and propose a novel approach that promises high classification accuracy by combining previously unused features with time-aware deep learning strategies.

Figures (6)

• Figure 1: Original picture with RGB color channels: Red (left), Green (center), Blue (right).
• Figure 2: HSV color channels with corresponding histograms: Hue (left), Saturation (center), Value (right).
• Figure 3: $\text{Y'C}_\text{b}\text{C}_\text{r}$ color channels with corresponding histograms: Luma (left), Chrominance Blue (center), Chrominance Red (right).
• Figure 4: neighborhood of center pixel $c$ (marked with round disk) with $P=8$ pixels for different radius values ($R$). The representation (black and white box) of pixel $p$ is filled black, if $\delta(g_p-g_c)=1$.
• Figure 5: Typical patterns discovered with .