Research On The Key Technologies Of Snapshot Multi-dimensional Imaging

Imaging techniques record the information of incoming light rays.Photons carry abundant information which can be denoted in nine dimensions(x,y,z,?,?,?,t,?,?).This nine-dimensional information gives the spatial(x,y,z)and angular(?,?)coordinates,wavelength(?),time(t),polarization orientation and ellipticity angles(?,?)of an incoming light ray.Snapshot multi-dimensional imaging is a type of technique that can derive the multi-dimensional information of a target scene in a single shot.Thanks to its snapshot multi-dimensional imaging capability,this technique has become an important direction in the field of imaging and been wildly applied in biomedicine,environment monitoring,target recognition,military surveillance,and so onCapturing the high-resolution high-dimensional information of a target scene is one of the important directions in the field of imaging.Limited by the amount of raw data captured by the detector in a single integration time,snapshot multi-dimensional imaging techniques suffer a serious trade-off problem between the number of dimensions and the resolution in each dimension they can achieve.In recent years,researchers introduced compressed sensing into multi-dimensional imaging and mitigated the trade-off problem to a certain extent.However,the multi-dimensional imaging techniques based on compressed sensing is computationally expensive and highly relies on the ill-posed assumption that the multi-dimensional information is sparse in a given domain,which significantly limits their applications.Using a detector array can increase the amount of raw data acquired by the imaging system in a single shot,however,the multi-dimensional imaging techniques based on detector array are bulky,unstable and expensive.Therefore,it is an urgent and significant subject for the development of snapshot multi-imaging to achieve the compatibility of high-dimension,and high-resolution.This paper aims to solve the above problems and provides key technology and theoretical basis for snapshot high-dimensional high-resolution imaging.The research findings can be applied in biomedicine,environment monitoring,and target recognition.The main contents of this paper include:(1)To solve the problem that snapshot spectral imaging techniques cannot achieve high-resolution in both spatial and spectral domains,this paper proposed a high-resolution snapshot imaging spectrometer based on spectral-panchromatic co-aperture imaging.This method uses two different information-retrieval strategies:Fourier transform spectral imaging and panchromatic imaging.Meanwhile,this paper carried out the studies on the fusion of spectral datacube and panchromatic image.Based on the sparsity of the spectral information,this paper designed an image fusion algorithm based on grouping principal component analysis(GPCA)to suppress the spectral distortion during the image fusion.The experimental results demonstrate that the prototype can increase the spatial resolution of the reconstructed spectral datacube from 0.8911p/mm to 3.56 1p/mm.(2)To achieve the compatibility of high-dimension and high-resolution,this paper proposed a snapshot spectral-light-field imaging method.Base on the sparsity of the objective information,this method introduces focused light-field imaging into the snapshot hyperspectral imaging Fourier transform spectrometer and captures the raw image containing both the light-field and interference of the target scene using a single detector.On the premise of maintaining the accuracy and resolution of the reconstruct data,this method overcomes the conflict among compactness,high-dimension,and high-resolution.To our knowledge,this is the first single-detector imaging system to record the seven-dimensional plenoptic function of light rays from an object.The experimental results demonstrate that the imaging system can derive the seven-dimensional plenoptic function of light rays from dynamic objectives using a single detector.(3)To overcome the limitation of the traditional presentation attack detection(PAD)methods in face recognition,this paper proposed a face PAD method based on snapshot multi-dimensional imaging.Combining the differences between real faces and presentation attack instruments on spectrum and three-dimensional profile,this method trains two support vector machine classifiers based on the spectral datacube(x,y,?)and light field datacube(x,y,?,?),respectively.A decision-level fusion algorithm is used to combine the classification results of these two classifiers and dramatically improves the performance of the proposed PAD method.The experimental results demonstrate that the performance of the PAD method based on snapshot spectral-light-field imaging is better than the state-of-art techniques.