Images,which serve as a principal carrier of information,have drawn the attention of researchers due to the emerging technologies in compression and reconstruction techniques.When the theory of compressed sensing,which is used for image signal compression and reconstruction,is applied,it requires an observation matrix that is large enough,an aspect that restricts real-time sensing.The introduction of block compressed sensing algorithms has overcome the drawbacks,such as prolonged reconstruction time and high computational complexity,which are found in compressed sensing reconstruction.However,block compressed sensing,which applies the same sampling rate to different image blocks,tends to overlook the differences in texture information between blocks.This results in an irrational distribution of sampling resources.This paper is centered around this issue and investigates the block compressed sensing algorithm for images,with the following work:(1)An analysis was conducted,which involved studying the impacts of four different block sizes and four reconstruction algorithms on the block compressed sensing reconstruction performance,under the condition of uniform segmentation.Using the combination of a 32x32 blocks and the SPL reconstruction algorithm as the condition,it was clear that the conventional block compressed sensing algorithm suffered from a limitation-the usage of the same sampling rate.To tackle this,an algorithm was proposed,an algorithm that was based on one-dimensional gray entropy for image-adaptive block compressed sensing.This novel algorithm had a unique capability,which was the ability to characterize the texture information of an image through one-dimensional gray entropy.This was a feature that allowed the algorithm to adaptively assign the sampling rate and conduct the reconstruction in line with the richness of texture information present in each image block.When the results of the experiments,which were conducted in the past,were scrutinized,it was clear that the proposed algorithm had better reconstruction performance,a performance that was superior compared to the block compressed sensing algorithm that used a fixed sampling rate.(2)Addressing the issue of poor reconstruction quality of image blocks,which was due to the limited texture characterization capabilities of a single image feature in adaptive block compressed sensing algorithms,this paper presented a texture information-based image adaptive block compressed sensing algorithm.At the observation end,the one-dimensional gray entropy and the standard deviation,which were combined,had been used to measure the richness of texture information in image blocks.The K-means++ clustering algorithm,which was introduced,classified the image blocks into three categories based on texture similarity.The edge information,when combined,allowed adaptive sampling rates to be assigned to each category of image blocks.This resulted in a significant reduction in the number of observation matrices while achieving a fine partitioning of sampling rates.Secondly,To alleviate the block effect that resulted from the fragmentation of correlations due to image partitioning,improved the uniform block structure.This was accomplished by proposing overlapping block structures and hierarchical block structures.After comparing the de-blocking performance of the three structures,a reconstruction based on the hierarchical block structure was chosen.This structure had a block observation-merge reconstruction at its core and was complemented by the SPL algorithm.Experimental results showed that,when compared to block compressed sensing algorithms and adaptive block compressed sensing algorithms,the proposed algorithm effectively alleviated the block effect in reconstructed image blocks.It also demonstrated superior reconstruction performance and shorter reconstruction time.(3)Considering the underutilization of high and low frequency information from wavelet transform in multiscale block compressed sensing algorithms,a new image adaptive multiscale block compressed sensing algorithm was proposed.This algorithm,which combined dual-domain information and directional features,introduced a weighting factor that was based on multiscale block compressed sensing.This factor weighted the combination of the spatial total variation of low-frequency approximation image blocks and the high-frequency coefficient values of the corresponding blocks in the wavelet domain,providing guidance for the pre-allocation of sampling rates for high-frequency coefficient blocks.Secondly,the directionality of the image was estimated,an important process that acknowledged the distinctive feature of differing texture directions among high-frequency sub-bands.Consequently,the algorithm was able to fine-tune the allocation of sampling rates for high-frequency coefficient blocks according to the principal directions of each sub-band,an adverbial clause showing how these adjustments were made.Finally,considering the fact that multiscale block compressed sensing algorithms had not dealt with low-frequency approximation images,the researchers of the paper introduced a spatially adaptive weighted filter.This filter,which was designed to reconstruct low-frequency approximation images,protected edge details while alleviating jagged block effects.To reduce reconstruction time,which was a crucial aspect,a hierarchical block structure,combined with the SPL algorithm was used for high-frequency coefficient reconstruction.The experimental results showed that the quality of the reconstructed images using the proposed algorithm had seen certain improvements,and the edge block effect was alleviated,a relative clause highlighting a key outcome of the study. |