Research On Image Adaptive Multi-scale Block Compressive Sensing Algorithm

In practical applications,it is usually difficult to obtain high-resolution images due to the hardware device or the bad environment for acquiring images.Obtaining high-resolution images by improving the performance of hardware devices significantly increases the cost of acquiring images.The super-resolution reconstruction algorithm of an image is based on the existing hardware system and performs learning or interpolating or modeling in one or more low-resolution images and a higher-resolution image is obtained through reconstruction.When the above three methods are used to reconstruct an image,the requirements of storage space and transmission bandwidth are higher.Reconstruction algorithm based on sparse representation comes into being,which applies the compressed sensing algorithm to single image super-resolution reconstruction.Compressive sensing algorithm firstly overcomes the limitation of 2 times sampling of Nyquist sampling theoretically.Secondly,it simultaneously carries observation and compression at the encoder end,which effectively reduces the observation time and running time of the sensor.The application of compressed sensing involves many fields such as pattern recognition,signal and image processing.This paper mainly studies the application of block-based compressed sensing algorithm in natural image super-resolution reconstruction.The image super-resolution reconstruction algorithm based on block compression sensing cuts the image into blocks of the same size and performs observation reconstruction on each sub-block separately,which effectively reduces the calculation cost and storage space.The multi-scale block compression sensing algorithm performs wavelet transform on the original image,and each block adopt the different sampling rate,so that the image reconstruction quality can be improved.However,because the multi-scale block compression sensing algorithm ignores the role of high-frequency signals in reconstruction,a large amount of edge and detail information is lost.In order to further improve the effect of image reconstruction,this article focuses on the importance of high-frequency information in the reconstruction conducts the following two research work.Firstly,in the process of how to effectively use high-frequency information,an adaptive multi-scale block-slice compression sensing algorithm is proposed.The algorithm firstly performs N-layer wavelet transform to obtain a low-frequency signal and a high-frequency signal,which are respectively divided into non-overlapping blocks of the same size after wavelet inverse transform,and 2-dimensional neighboring edge adaptive weighted filtering is applied to the low-frequency part.In this processing,the block effect of the reconstructed image is effectively eliminated by the smoothing filter.A texture-adaptive block-sampling is applied to the high-frequency part,and an adaptive sample and observation is also conducted according to the amount of information contained in the image sub-block.This algorithm makes full use of the high-frequency signal which obtained by wavelet decomposition so that allows the edges of the image can be effectively reconstructed.Secondly,after studying the adaptive block size reconstruction algorithm,which is soon after applied to the reconstruction of high-frequency signals,an adaptive multi-scale block size compression sensing algorithm is proposed.Base on decomposed high and low frequency which are processed by the adaptive multi-scale block-compressed sensing algorithm,when high-frequency signals are processed,the adaptive block size is sampled according to the information entropy of the image,Larger blocks are used for sampling observations for smooth blocks containing less information,and smaller blocks are used for sampling observations for blocks containing complex texture informationCompared with existing block compression sensing algorithms,edge-and-direction block-based compression-sensing algorithms,and block-based compressed-sensing algorithms based on texture and direction,the performance of this algorithm has been improved at different sampling rates.The high-frequency signal which represents the detailed information is fully reconstructed and the sampling time is effectively reduced.After the two-dimensional neighbor edge adaptive weighted filtering algorithm is processed,the block effect of the reconstructed image can be effectively removed.The reconstructed image obtained by the improved algorithm has a higher resolution,especially for the more detailed image reconstruction,a higher peak signal-to-noise ratio is obtained.