Research On Adaptive Measurement Methods For Block Compressive Sensing Of Images

The traditional image sampling method with the basic framework of Nyquist sampling theorem puts high cost on the sensor,and it is not applicable in practical applications when the acquisition time,memory consumption,or computing speed is limited.Compressed sensing theory shows that the signal having sparse or compressive representation can be exactly reconstructed with high probability,which breaks through the limitations of the Nyquist sampling theorem and widely used in the field of image coding.Aiming at the adaptive measurement method of block compressed sensing,after reviewing the related research background and basic theoretical knowledge,the measurement structure of adaptive block compressed sensing is firstly analyzed.Then,based on the measurement structure,image features that measure block sparsity in the pixel domain is proposed and adaptive measurement methods are also designed.Finally,considering that original image is not available in Compressive Imaging applications,an adaptive measurement method is performed in the measurement domain.The main contributions of this dissertation are as follows:(1)The three types of block compressed sensing measurement structures are proposed.Image blocks are the basic unit in the process of image measurement and reconstruction.A good block measurement structure can effectively reduce the blocking artifacts and blurs.First,a raster structure is proposed.In this structure,two-dimensional images are compiled in columns and they are stored and measured respectively.Second,considering that the blocking artifacts usually results from the obvious differences in the sparsity of adjacent image blocks,a patch structure is proposed.The two-dimensional image is divided into overlapping image blocks and raster scanned into one-dimensional vectors,which can effectively smoothen the adjacent blocks.Finally,because there are lower computational complexity in the smaller blocks and it is easy to measure them,while the quality of reconstruction can be improved in the larger blocks,a layered structure is proposed.The layered structure divides the original image twice.The original image is divided into several non-overlapping larger blocks firstly,then the larger image blocks are divided into smaller blocks.This can effectively guarantee that small blocks are measured and large block are reconstructed,which ensures the reconstruction quality of the original image.(2)The two features in pixel domain are proposed to measure the sparsity of image blocks.In order to reduce the blocking artifacts,the concept of spatial entropy is proposed to reflect the block sparsity.In each block,measurement times can be allocated adaptively by using spatial entropy,which reduce blocking artifacts effectively after image reconstruction.Because there are correlation between image blocks,the visual contrast is calculated.It is used for reflecting the block sparsity and measurement times of each block are allocated adaptively.The visual contrast between blocks can reflect the sparsity distribution of the blocks,thereby the blocking artifact is reduced.(3)Sensed entropy-assisted adaptive measurement method is proposed.Aiming at the unavailability of original image in the encoding end constructed by the Compressive Imaging devices,an Entropy-Assisted adaptive measurement method is proposed.By using the compressed sensing measurements,the sensed entropy of each image block can be estimate directly.Through the sensed entropy,the sparsity of the image block are measured and the adaptive distribution scheme of measurement times is designed.This method can reduce redundant measurement values and shorten measurement time effectively.Compared with the features in pixel domain,sensed entropy can still well reflect the distribution of block sparsity.By using the sensed entropy,the measurements of redundancy are reduced and the blocking artifact can be inhibited significantly,thus the visual quality of reconstructed images is improved effectively.