A novel method to enhance color spatial feature extraction using evolutionary time-frequency decomposition for presentation-attack detection

Vulnerability to presentation attacks is the most valid issue of face-based authentication systems. Therefore, automatic detection of face spoofing plays a vital role in the safe use of face recognition applications in situations where the system works alone. In this work, we propose a method based on texture feature analysis. We select varying color channels among RGB, HSV, and YCbCr spaces depending on the minimum classification error rate to extract different wavelet sub-bands. Accordingly, the Green (G) channel of the RGB color spaces, the Saturation (S) of the HSV color space, the blue-difference Chroma (Cb) component of the YCbCr color space, and the Grayscale of the facial image to extract wavelet sub-band. The final texture feature vector was constructed using the Local Phase Quantization (LPQ) descriptor on the obtained wavelet sub-bands. Moreover, we use the genetic algorithm to reduce the feature vector’s dimensions and minimize the classification error rate. The proposed method’s performance was evaluated using inter-dataset and intra-dataset tests on nine public datasets. In these tests, the performance of the proposed method on 3Ddmad, HKBU-MARsV1+, Replay-Mobile, OULU, SiW, WMCA, CASIA-MFS, and MSU-MFSD datasets has proven to be better than the most advanced methods available.