A Sampling Rate Adaptive Block Compressed Sensing Method Combining Gray Entropy and Blind Deconvolution

Abstract

The invention discloses a sampling rate self-adaptive block compression sensing method combined with gray entropy and blind deconvolution. First, the image is subjected to three-layer wavelet transformation, and the gray entropy of each image block is obtained by sub-blocking of low frequency subbands. , and then use the gray entropy as the judgment condition to adaptively allocate the sampling rate to each block of the high-frequency sub-band, assign more sampling numbers to complex texture blocks, and lower sampling numbers to simple smooth blocks, and finely process human Eye-sensitive texture blocks, efficiently process smooth blocks with low attention, and perform blind deconvolution processing on low-frequency sub-bands; then, compress and sample high-frequency sub-bands based on the obtained adaptive sampling rate, and use SPL The method fully reconstructs the details of each block. Finally, the low-frequency sub-band image after blind deconvolution and the high-frequency sub-band image after compression and reconstruction are normalized and superimposed to obtain the restored image for transmission. The invention reduces the iterative complexity and improves the accuracy and completeness of reconstruction details.

Description

technical field [0001] The invention belongs to the technical field of compressed sensing, and in particular relates to a sampling rate adaptive block compressed sensing method combined with gray entropy and blind deconvolution. Background technique [0002] Compressed Sensing (CS) theory shows that if a signal is sparse or sparse under a certain basis, then the sparse signal can be multiplied by an irrelevant observation matrix for dimensionality reduction observation, and obtain Fewer observations (sampling values), and finally obtain the reconstructed original signal by solving an optimization problem. However, in the process of directly applying CS theory to image processing, if the entire image is directly observed, the required observation matrix is ​​often on the order of 104-106, which undoubtedly increases the storage and transmission burden again. Block Compressed Sensing (BCS) theory is a good application of CS theory to the field of image processing. This metho...

AI Technical Summary

Problems solved by technology

[0004] 1. The existing BCS-SPL method based on wavelet transform performs wavelet transform on all image blocks one by one. The number of iterations of the method is large, and the integrity of the edge details of the high-frequency sub-bands of the image is not considered. The low-frequency sub-bands of different image blocks are in the Due to the irregular frequency jump at the edge of the block in the superimposed backfill at the receiving end, the block effect is also aggravated

[0005] 2. The existing BCS-SPL method based on wavelet transform uses the same sampling rate for fixed observation, and does not allocate the sampling rate to each image block according to the difference in texture structure or sparsity, resulting in undersampling of texture blocks and smooth blocks. Oversampling, weakening method performance

The existing BCS-SPL method based on wavelet transform does not make full use of the prior information that can reflect the structural characteristics of the image, only the low-frequency sub-bands are reserved, and the high-frequency signal after compression and reconstruction is simply normalized overlay, resulting in waste of low-frequency sub-band information resources

Images