Ghosting Reflection Compensation for Multispectral High Dynamic Range Imaging

A compact overall design of a multispectral camera can be achieved by placing the bandpass filters between the lens and sensor. However, internal reflections between the sensor (or IR cut filter) and the optical filters may lead to weak duplicate images which impair the image quality. The duplicates called ghosting are especially noticeable near bright image regions and interfere with the surrounding image content. When using combined high dynamic range (HDR) and multispectral imaging, the increased dynamic range is of reduced value due to the ghosting. In the current paper, we model the ghosting effect and present a calibration setup for the characterization of the ghosting, which utilizes a backlit film calibration pattern. We perform HDR imaging to acquire the vast dynamic range of the backlit scene. Our calibration algorithm then estimates the model parameters and allows a compensation of the ghosting. As long as the lens parameters are not changed, the calibration may be used for all subsequent images. We give detailed results for the calibration and the effect of the compensation.