Modeling and Compensation of Ghosting in Multispectral Filter Wheel Cameras

Multispectral filter wheel cameras divide the visible electromagnetic spectrum by using several optical bandpass filters mounted on a filter wheel and acquire one color component for each filter wheel position. Afterwards, the single images are combined into one multispectral image. While the color accuracy of this approach and the stop-band attenuation of the bandpass filters is superior to other technologies, ghosting images are produced by reflections between the image sensor and the filter surface: The original image is duplicated in a displaced, weaker and softened form and added to the original image, thus compromising the original. We analyze the path of rays in this specific optical setup and derive a physical model for the ghosting reflections. By linking the physical model to the image content, we derive a calibration and compensation algorithm, whose parameters are estimated from a test image. Application of our correction algorithm makes the ghosting virtually vanish.