Research On Core Image Compression And Reconstruction Based On Compressed Sensing

Core image plays an important role in the exploration and research of oil and gas fields.Because of the special and complicated geological evolution and geological environment, the surface of core images are complex in edge, rich in texture and great different with the characteristics of natural images. With the core image acquisition equipment widely to be put into use, the amount of the core image data of domestic oil fields which collected by such equipment is also increasing, which brings great difficulties to the storage and transmission of core images. Under these circumstances, high efficient acquisition, compression and reconstruction of core images with mass data is the key to improve the utilization of core data.Therefore, this thesis combines the theory of wavelet analysis with that of compressed sensing,and proposes a reasonable algorithm to compress and reconstruct core images, which can improve the compression effect and visual quality. The main research contents of this thesis are as follows:(1) A K-SVD with the adaptive threshold compressed sensing reconstruction of core image is put forward. Aimed at the partial detail loss problem of compressed sensing reconstruction of core images, according to sparse representation and the characteristics of core images, the method conducts block for core images, obtains observations of image blocks by Gaussian random matrix, and then calculates the threshold through the initial solution of core image reconstruction. The reconstruction can be achieved by Wiener filtering combined with Landweber iterative, which also combined with K-SVD dictionary. The experimental results show that the dictionary can reserve more image details, and the algorithm achieves a high quality reconstruction of core images.(2) A multiscale adaptive compressed sensing reconstruction of core image using information entropy is put forward. Aimed at the detail vague problem of core images reconstruction using Block Compressed Sensing-Smooth Projected Landweber, according to the core image texture and the human visual sensitivity, the method introduces discrete wavelet transform into the sparse representation, conducts multiscale block for each subband, and adaptively allocates the sampling rates and determines the measurement matrix. The reconstruction can be achieved by Wiener filter combined with Landweber iterative. The experimental results show that the algorithm preserves the texture features of core images effectively, achieves good subjective visual quality and improves the PSNR value by 2-4dB, compared with that of Block Compressed Sensing-Smooth ProjectedLandweber algorithm under the same sampling rates.(3) A core image observation SPECK compression based on block compressed sensing is put forward. Aimed at the problem that using the existing compressed sensing to process core images, the effect of high compression ratio is not ideal, according to the characteristics of core images and compressed sensing, the method introduces discrete wavelet transform into the sparse representation and conducts multiscale block for each subband, and then allocates different sampling rates to different levels. The image block observations can be achieved by the corresponding level measurement matrix. Compared DPCM with SPECK to compress observation of core images. The compression and reconstruction of core images can be achieved by Wiener filter combined with Landweber iterative. The experimental results show that the proposed algorithm improves PSNR of core images reconstruction, compared with that of multiscale block compressed sensing-based DPCM plus uniform scalar quantization and uniform scalar quantization under the high compression ratio.