Research On Adaptive Compressed Sampling And Image Reconstruction Method Based On Extended Wavelet Tree

Limited by the Nyquist-Shannon sampling theorem and the compressing-after-sampling framework,the traditional image sampling technique inevitably samples a fair amount of redundant information,leading to a waste of sampling resources.The digital imaging technique based on Compressive Sensing(CS)theorem exploits the sparsity of natural images,combining sampling and compression,which improves the utilization efficiency of samples and lowers the hardware requirement for sampling system.However,the CS based imaging technique has shortcomings in imaging quality and computation complexity of image reconstruction,and these shortcomings become more evident with the continuously growing needs for imaging resolution.To solve these problems,this thesis studies an adaptive compressed image sampling and reconstruction approach based on CS and extended wavelet tree.The proposed approach can lower the algorithm complexity of reconstruction,shortening the time of reconstruction,and ameliorate the issues in compressed sampling and reconstruction of high-resolution images such as noise,blocky effect,and color distortion,improving the quality of imaging.The main research contents are as follows.To shorten the computation time of image reconstruction and lower the algorithm complexity in CS based optical imaging,we propose an adaptive compressed imaging approach based on subdivision-controlled Digital Micromirror Device(DMD)and extended wavelet tree structure.The proposed approach applies the subdivision-controlled DMD to the sampling based on extended wavelet tree theory,which dramatically reduces the storage space for the measurement matrix and improves the sampling efficiency.The experimental results illustrate that the proposed approach can achieve high quality imaging with low sampling rate.This thesis proposes an adaptive compressed imaging technique based on Hadamard coded measurement.We form the DMD sampling patterns using the Hadamard matrix code which measures and samples the corresponding space region in blocks.The experimental results demonstrate that the proposed method can effectively reduce the noise interference and improve the Peak Signal-to-Noise Ratio(PSNR)of imaging with less amount of sampling data.The proposed technique is suitable for high sensitive imaging under the condition of weak light signal.Since the adaptive compressedimagling approach based on wavelet tree structure only samples important wavelet coefficients,there are blocky effects in the restructured images.To remove the blocky effect,we propose an adaptive compressed imaging approach based on wavelet domain interpolation.We not only sample the important wavelet coefficients in each layer of the wavelet tree adaptively,but also estimate the unsampled unimportant wavelet coefficients with wavelet domain interpolation to improve the quality of imaging.The experiments show that the proposed approach obtains smoother and higher quality images in both noisy and noiseless circumstances.To overcome the low time-effectiveness and color distortion in the sampling of color images,we propose an adaptive compressedimaging approach for color images based on Multitask Bayes model.Based on the correlation among the red,green,and blue channels,we perform inverse wavelet transformation on the processed wavelet coefficients to obtain the red,green,and blue gray scale component images respectively and the final reconstructed image through fusion using CS based on Multitask Bayes model.The experimental results demonstrate that the Multitask Bayes model can reduce the color distortion,and hence guarantee the color consistency and quality of the reconstructed color images.