Exploiting Feature-Based Fusion in LED-based Multi-Spectral Imaging

Multi-spectral recording systems are used in numerous applications ranging from quality assurance over biometrics to remote sensing. This paper reports on the feasibility of a LED-based multi-spectral imaging system where the spectral characteristics of the illumination is changed by activating different LEDs. The multi-spectral images are captured by a cost-efficient CCD camera. We focus here on applying methods from sensor fusion to increase the classification performance. Various features are generated from the different channels of the multi-spectral images. The most discriminative features are then selected by a forward selection strategy. We demonstrate our feature-based approach in human vein detection. Various test data have been recorded by our prototype of the LED-based multi-spectral capturing system and have served as basis for the experimental evaluation. A detection performance of at least 96 % has been achieved.