Research On Adaptive Compressive Sampling Imaging Based On Wavelet Transformation

The traditional information sampling process must follow the Nyquist sampling theorem that the sampling bandwidth is more than twice of the frequency bandwidth,which can cause the increase of transmission costs and the waste of storage resources.Compressed sensing theory breaks the limitation of Nyquist sampling theorem,by combining the compression and sampling of the signal,the transmission efficiency of the image signal is greatly improved,and the high resolution image signal can be obtained under the low cost condition.At present,an imaging device called single-pixel camera has been developed based on the compressed sensing principle.The imaging principle of the single-pixel camera is to use a single detector and digital micro-mirror device DMD to sample and measure the image,and then to reconstructe the sampled target image information.The main goal of the single pixel imaging is to use the limited sampling resources to obtain the target image information as much as possible.By using the adaptive compressive sampling method,the main feature information of the compressible image signal can be adaptively sampled and marked,which can reduce the consumption of sampling resources and reconstruct the high-quality images.For how to determine the key sampling area of the target image,and how to allocate the limited sampling resources to obtain high-quality images is the main direction for this study of adaptive compressive imaging.Considering that the image under wavelet transform has the feature of sparse,the adaptive compressive sampling imaging of wavelet transform is studied as follows.1.In the wavelet tree structure formed by the wavelet transform several times,the adjacent high-frequency feature coefficients has the parent-child relationship between the layers and the sibing relationship within the layers.Using the correlation between the neighborhood coefficients of the wavelet tree,this thesis is designed an adaptive compressive sampling algorithm based on wavelet tree correlation coefficient importance prediction.The algorithm uses the correlation coefficient of the adjacent parent coefficients as the condition to predict the important child coefficients,after the prediction is completed,the coefficients obtained by the sampling are inversely transformed by the wavelet to reconstruct the image.Through the analysis and verification of the simulation results,the algorithm designed in this thesis can reconstruct high quality images,compared with other algorithms,the quality of the reconstructed image with the same sampling rate are further improved.In the low noise environment,the noise immunity of this algorithm also has a good performance.2.An adaptive compressive imaging algorithm based on wavelet prediction is analyzed,which can predict and sample large coefficients by distributing sampling ratio of different coefficient layers.Before obtaining the wavelet coefficients of each prediction,it is necessary to interpolate rough approximate images with high resolution reconstruction.In this thesis,through experimental comparison,it is found that different interpolation algorithms have different degrees of influence on the quality of the final reconstructed image.Therefore,a scheme of alternately using interpolation algorithm is designed,which determines the optimal interpolation algorithm by comparing the threshold with the sampling ratio of each layer.By comparing the experimental results,the improved scheme in this thesis has the effect of improving the reconstruction quality of the image.For the same compression ratio,this thesis reassigns the sampling ratio of the last two layers,and analyzes the effect of different sampling ratios on the reconstructed image through experiments.