A CTIS Image Reconstruction Method Based On Low-rank Constraint

Hyperspectral imaging technology is a new multidisciplinary remote sensing technology capturing spatial and spectral information of a scene,which has been applied in many fields such as airborne sensing,biomedical imaging and space surveillance.Specially,the snapshot hyperspectral imaging can capture the real-time scenes because of its extremely short collecting time.Image reconstruction is a key problem in snapshot hyperspectral imaging.In this paper,we propose an image reconstruction algorithm of computed tomography imaging spectrometry(CTIS)based on the intrinsic characteristics of hyperspectral images.The main contents of this dissertation are as follows:1.CTIS image acquisition: A series of hyperspectral data has been collected by the self-building CTIS system in our laboratory.2.CTIS image preprocessing:(i)The background removing via subtracting the system noise.(ii)The high dynamic range is achieved using the multi exposure image fusion method.(iii)The raw images are demosaiced by a weighted edge adaptive algorithm.3.CTIS image reconstruction: The low rank property of hyperspectral images is coming from the spatial correlation and spectral correlation of hyperspectral images.The low rank constraint of image reconstruction is established by the low rank matrix approximation.Combing with EM iterative method,a low rank constrainted hyperspectral image reconstruction is successfully applied in our CTIS system.Based on low rank constraint,our proposed CTIS image reconstruction algorithm can effectively reconstruct the hyperspectral data cube,which also can preferably suppress noise.The peak signal to noise ratio(PSNR)of the reconstructed image can be up to 66 dB,which is improved by 6 dB compared with traditional method.The spectral angle mapping(SAM)is 0.041,which is 4 times smaller than that of the traditional method.