Research On Enhanced Block Compressed Sensing Of Images Based On Total Variation Using Texture Information

The compressed sensing is so different from traditional theorems that it breaks the limit of Nyquist sampling theorem and can use a low sampling rate to construct original image if signals are sparse or compressible. Because compressed sensing could greatly reduce the valve of sampling required,it can solve the problem of how to save a compressing videos and photos while receiving them. The compressed sensing theory contain three main parts including:Sparse representation,mainly about the choice of measurement matrix;sampling, mainly about the generate of observation vector; reconstruction, about how to regain the original image from the observation vector. Reconstruction is the most significant part. It directly determine the final result of reconstructed image.To reconstruct an image in compressed sensing is to recover a sampled signal from lower dimension to its original high dimension. Its mathematical model is an underdetermined equation. We focus on adaptive block-based compressed sensing and find a way to combine TV filter with compressed sensing. We have mainly done works as follows:(1)This thesis propose an adaptive block-based compressed sensing model which changes the sampling rate of each block instead of a constant one by using texture information. This model will improve the effect of reconstructed image by properly allocate sampling rates of each block and has a low computational complexity and good robustness.(2)The proposed algorithm join TV filter with SPL process,and build a DDWT/TV filter model based on texture information to replace the former filtering process in reconstruction. The model could preserve more details of the image after the decreasing of block artifacts by using adaptive sampling,and avoid the problem that the Wiener filter may not be suitable for an image whose block artifacts are weak.