This paper presents a novel Discrete Cosine Features (DCF) method for face recognition.
The DCF method works by fusing the complementary features derived from the Discrete Cosine Transform (DCT) of the color component images in the YIQ color space. The novelty of the DCF method thus comes from both the multiple imaging (three component images) in the YIQ color space, and the multiple face encoding (different masking) in the DCT frequency domain. First, each color component image in the YIQ color space is transformed to the frequency domain via DCT, where three DCT frequency sets are derived by means of masking to encode the image at different representation levels (the reconstructed images display different details). Second, the three DCT frequency sets at the same representation level across the Y, I, and Q color component images are concatenated—the feature level fusion—to form an augmented pattern vector. Third, the complementary features from each of the three augmented pattern vectors (corresponding to the three different representation levels) are extracted using an Enhanced Fisher Model (EFM). Finally, the three similarity matrices generated using the complementary features are fused by means of the sum rule—the decision level fusion—to derive the final similarity matrix for face recognition. The effectiveness of the proposed DCF method is demonstrated using a complex grand challenge face recognition problem and a large-scale database. In particular, the Face Recognition Grand Challenge (FRGC) and the Biometric Experimentation Environment (BEE) show that for the most challenging FRGC version 2 Experiment 4, which contains 12,776 training images, 16,028 controlled target images, and 8014 uncontrolled query images, the DCF method achieves the face verification rate (ROC III) of 81.34% at the false accept rate of 0.1%, compared to the FRGC baseline algorithm face verification rate of 11.86% at the same false accept rate. (Published abstract provided.)