Image Compressed Sensing Coding Techniques Based On Discrete Wavelet Transform

The compressed sensing(CS)theory has broken through the limitation of the Nyquist sampling theory,it can compress the signal while sampling the signal and it's applied to the image coding field.Thesis has studied the constructed of sparse vector and measurement matrix in image compressed sensing coding algorithm,and proposed three different coding algorithms,The main innovations are as follows:(1)A compressed sensing coding algorithm is proposed based on image multi-layer wavelet full sub-band.At the coder,decomposition the image with the multi-layer wavelet transform,constructed sparse vector with all wavelet transform sub-bands,that is constructed sparse vector from a lowfrequency coefficient,a highest level high-frequency coefficient and the descendant coefficients.Designed weight matrix according the different importance of high-frequency sub-bands on image reconstructed,to improve the measurement matrix,then compressed sensing the sparse vector.At the decoder,extracted the first coefficient of the reconstructed sparse vector,then obtain the lowfrequency by mean processing and reduction all the high-frequency sub_bands,finally reconstructed with multi-layer inverse wavelet transform to obtain the image.(2)A block wavelet transform compressed sensing coding algorithm is proposed based on local importance.The algorithm uses the correlation between image block variance and texure and human Region of Interest(ROI)as parameters of local importance.At the coder,divided the image into blocks,which have the same size and non-overlapping,then divided the blocks uniformity into k classes according to the size of variance and ROI,set k measurement rates combined with the overall image requirements;decomposition the image by wavelet transform,then preserve the low-frequency subbands,and reshape the block of high-frequency sub-bands into one-dimensional sparse vector,select corresponding measurement rate according to the region of block,finally compressed sensing the one-dimensional vector.At the decoder,reconstructed the blocks by inverse wavelet transform of reconstructed high-frequency sub-bands and low frequency sub-bands,finally combined all the blocks to obtain the image.(3)A full sub-bands compressed sensing coding algorithm is proposed based on wavelet packet decomposition.The algorithm uses the wavelet packet decomposition to sparsity image and completes the blocks at the same time.At the coder,decomposition the image by wavelet packet transform,constructed sparse vectors with wavelet packet full sub-bands,that is constructed a sparse vector from a coefficient with all sub-bands at the same position;then calculated the energy of each sub-band as the weight of sub-band,and form a weight matrix to improve the measurement matrix,finally compressed sensing the sparse vector.At the decoder,constructed the sparse vector,and reduction to wavelet packet coefficients matrix,finally wavelet packet reconstruction the coefficients matrix to obtain the image.In order to verify the performance of three algorithms,thesis carried on the detailed algorithm simulation and result analysis.It shows that the three algorithms have good subjective visual effects and PSNR.Finally,thesis summarizes the full text work,analyzed the improvement of three algorithms and future research direction.