Multispectral Imaging for Illumination Invariant Face Recognition

Under controlled illumination conditions, visual Face Recognition systems perform well for faces with low or no disguise [1]. Infrared (Thermal) Face Recognition provides an enticing alternative to Visible Face Recognition due to the relative insensitivity of IR imagery to illumination changes and disguise [2]. Image fusion is in fact a combination that extracts redundant and complementary information from different source images and fuses them into one image which should be more informative and therefore more useful for human or machine perception [3]. Motivated by the success of Multispectral Fusion in areas such as Remote Sensing where fused images are obtained with both high spatial and spectral resolutions, we expect to improve overall Face Recognition Performance (FRP) by also using Multispectral Fusion. Literature on Multispectral Image Fusion is first reviewed, followed by a brief evaluation of FRP of the commercial software, FaceIt® (version: 3.0.2), in the presence of illumination changes. Then, by choosing appropriate weights of Wavelet transformation coefficients, a novel pixel-level Wavelet-based Data Fusion method which can take advantages of both visible and thermal imagery is described to improve FRP. Finally, tests and analysis of indoor Multispectral IRIS (Imaging, Robotics, and Intelligent Systems laboratory) database are introduced. In future, the above Wavelet-based pixel-level Data Fusion is expected to extend to Multispectral Face Recognition.