Color channel encoding with NMF for face recognition

• Published in: 2004 International Conference on Image Processing, 2004. ICIP '04 Vol. 3; pp. 2007 - 2010 Vol. 3
• Main Authors: Rajapakse, M., Tan, J., Rajapakse, J.
• Format: Conference Proceeding
• Language: English
• Published: Piscataway NJ IEEE 2004
• Subjects: Applied sciences; Artificial intelligence; Color; Computer science; control theory; systems; Data preprocessing; Degradation; Exact sciences and technology; Face recognition; Feature extraction; Image converters; Image recognition; Information systems. Data bases; Layout; Lighting; Memory organisation. Data processing; Pattern recognition. Digital image processing. Computational geometry; Shape; Software; Image recognition; Grey level image; Face recognition; Image processing; Image databank; Matrix factorization; Autoregressive model; Regression analysis; Pattern recognition; Green function; Color image; Noisy image; Luminance; Non negative matrix; Vector space; Facies; Illumination; Feature extraction; Robustness; Pattern extraction; Facial expression
• ISBN: 0780385543; 9780780385542
• ISSN: 1522-4880
• DOI: 10.1109/ICIP.2004.1421476

Colors act as cues for perceiving objects, particularly in complex scenes. Intuitively, color seems to play an important role in recognizing people in scenes. Recent research has evinced that color cues contribute in recognizing faces, especially when shape cues of the images are degraded. Although the input to many of the face recognition systems is color images, during preprocessing, these images are converted to gray scale images for the feature extraction. In this study, we use non negative matrix factorization (NMF) to recognize color face images. By using NMF, we encode color channels (red, green, blue), thereby, projecting these feature vectors to sparse subspaces. The implemented system is tested on a subset of color images in the AR database for robustness against facial expressions and illumination variations. Furthermore, color face recognition results are compared with the results obtained for the gray scale images of the same data set. Our results show improved accuracy of color image recognition over gray level image recognition when large facial expressions and illumination variations are present.