PROCEEDINGS ARTICLE | March 21, 2005
Proc. SPIE. 5681, Security, Steganography, and Watermarking of Multimedia Contents VII
KEYWORDS: Optical filters, Facial recognition systems, Detection and tracking algorithms, Cameras, Databases, Fourier transforms, Light sources and illumination, Image filtering, Optimal filtering, RGB color model
Face image recognition has been an active research topic for many years, and many algorithms have been proposed for the grayscale face images. However, only a few methods are devoted to color face images, even though most cameras produce color images. Color can be important for face recognition. Among the proposed color face recognition methods, most treat the three color channels separately and apply some grayscale face recognition methods to each of the channels and combine the results. In this paper we propose the quaternion correlation filter techniques for color face recognition by processing all color channels jointly. Quaternion numbers are the generalization of the complex numbers and can be considered as a number with a real part and three orthogonal imaginary parts. A color image (e.g., in RGB) may be represented using quaternion by encoding the three color components to the imaginary parts of the quaternion number. The quaternion correlation filter is extended from the standard correlation filter based on the recently developed concepts of quaternion Fourier transform (QFT), quaternion convolution (QCV), and quaternion correlation (QC). Face recognition is performed by cross-correlating (QC) an input face image with a designed filter. Numerical results show that quaternion correlation filters can improve the recognition performance over conventional face recognition methods.
