Research On Key Techniques Of Adaptive Compressed Imaging Based On Wavelet Sparse Sampling

The compressed imaging technology exploits the sparsity of natural scenes to sample and compress,which provides a novel way to solve the contradiction between the huge data sampling requirement and the limited detector resources.However,there are some shortcomings of the existing compressed imaging technology based on compressed sensing,such as uncontrollable imaging quality and a huge computation overhead of the reconstruction algorithms,which reduces the imaging speed and limits the scope of application.In order to build a fast and high resolution multidimensional compressed imaging system,the doctoral dissertation introduces a novel theory of compressed imaging,namely wavelet sparse sampling.To achieve adaptive imaging quality control,high efficiency measurement,and reconstruction with low computational complexity,the dissertation also studies on several key techniques of the wavelet sparse sampling,such as the compressed imaging algorithm and the compressed imaging system design,in the area of 2D imaging,video sequence acquisition,and 3D imaging.The main research work and contributions are summarized as the following aspects:To solve the existing problems of the compressed imaging method based on compressed sensing,the wavelet sparse sampling predicts the location of significant coefficients representing the sparse information in multiresolution,samples them with the corresponding wavelet basis patterns,and then retrieves the scene by performing an inverse wavelet transform with low computational complexity,based on the similarity between the wavelet transform and the human visual system.The theory can overcome the shortcomings of compressed sensing to achieve adaptive imaging quality control and high resolution real-time reconstruction.To improve the sampling efficiency of the significant wavelet coefficients,we extend the concept of wavelet trees by adding the sibling relationship to the wavelet trees structure,and then propose a single-pixel imaging technique,namely adptive compressed imaging technique based on Haar wavelet sparse sampling.The technique predicts the significant wavelet coefficients in groups along the extended wavelet trees,which effectively improves the predication accuracy.The number of measurements required for reconstruction can be significantly reduced by digging out the redundance in the DMD measurements.The compressed imaging system based on wavelet sparse sampling is built by adding a feedback mechanism on the single-pixel camera system.The simulation and experimental results show that the number of measurements is only 60%-70%of the measurements required for the method based on wavelet trees.In addition,for color imaging applications,we propose a colored adaptive compressed imaging technique based on wavelet sparse sampling in the YUV color space.By using the dependence on the brightness component and the chromaticity components in the YUV color space,as well as the principle of human visual system,the imaging time is significantly reduced while the image quality and color accuracy are improved.To achive fast and high-resolution video compressed imaging,an adaptive video compressed imaging technique based on wavelet sparse sampling is proposed.By constructing a multiresolution video compressed sampling framework,the technique alternately carries out wavelet sparse sampling and motion estimation,removing both the intra-frame redundancy and the inter-frame redundancy.The simulation and experimental results indicate that the sampling data can be reduced by 85%-95%,compared with the traditional digital video sampling technique.In addition,PSNR can be improved by more than 4dB by the proposed method compared with the CS-based video acuiqisiton methods such as 2DDWT and 3DDWT.To achive fast and high-resolution photon counting 3D imaging,we apply the theory of wavelet sparse sampling to the area of photon counting 3D imaging.First,a photon counting 3D compressed imaging system is built and a photon counting compressed sampling model is established.Then,since depth map is generally sparser than reflectivity,we propose an adaptive compressed photon counting 3D imaging technique based on the depth wavelet trees.Finally,a compressed photon counting 3D imaging technique based on depth compression and adaptive Hadamard basis scanning is proposed.The depth compression model is established to further reduce the computational complexity of reconstruction.The adaptive Hadamard basis scanning is employed to improve the photon collection efficiency,leading to better imaging quality in low light level.