Research On Hyperspectral Video Imaging System Using Amici Prism

The color of object is dependent on the spectral reflectance of it’s surface. Most of the imaging devices adopt RGB sensors to capture image. RGB sensors are the pixel units which response to R, G and B color respectively. When the device is capturing image, the light reflected by the object integrates over the three color channels respectively and the R、G、B color values of each pixel are captured finally. But these three color values are far from accurately representing the response of the observed object to various wave lengths and it can cause metamerism. The technology of spectral imaging system can capture object both from spatial domain and spectral domain. The system combines the function of imaging and spectral detection and it overcomes the metamerism.In order to solve the questions of low light collection efficiency and geometric distortion in the direct capture of hyperspectral video systems, the paper presents an advanced hyperspectral video imaging system using Amici prism. Compared to the traditional spectral imaging system, the proposed system’s light throughput gains up to 30% and the system is more compact with less geometric distortion. The key performances as a spectrometer are analyzed including spectral resolution, light throughput, spatial and temporal resolution. Additional experiments are presented on the light throughput.In consideration of the fake faces in the facial recognition such as photo, recorded video and high-emulation mask, the proposed advanced hyperspectral video imaging system using Amici prism is used for recognizing human face from high-emulation mask, based on the unique spectral reflectance ranging from 545 nm to 575 nm of the human skin. We choose appropriate spectral combination which makes the true and false face the biggest difference. Additional experiments are presented to validate the feasible of the scheme.In order to solve the question of capturing the video in the spectral imaging system based on filters, the paper presents multi-channel spectral imaging system using CMYG camera and full-multiband filter. The system can capture eight spectral images of the observed object in real time. Detailed theoretical analysis on the mathematical model and the physical map of the system is presented. Both direct and learning-based spectral reconstruction are used for the captured images. The reconstruction result of learning-based method is closer to the real spectral curve.