Research Of Underwater Imaging Technology And Image Compression Technology Based On Compressive Sensing Theory

Compressive Sensing (CS) is a new type of sampling theory. Compared with thetraditional Nyquist theorem, CS theory has many advantages, such as requiring fewersamples and less sampling resources. Recent results show that CS theory has greatpotential for many fields such as radar imaging, information processing and remotesensing. It is well known that underwater optical imaging has wide applications inoceanography, underwater scientific investigation, marine resource exploration andmilitary field, but the imaging distance and image quality of the existing underwateroptical imaging methods is hard to satisfy the actual demands. After deep research onthe superiority of CS theory on underwater imaging and image compression, someresearch results are taken in the dissertation as follows.The concept of underwater single-pixel camera is first put forward. Using theadvantages of range gating in eliminating stray light when imaging and ofunderwater laser pulse illumination in increasing imaging beam energy, the framestructure of underwater single-pixel imaging system based on laser pulseillumination is designed. According to the designed system, some software platformssuch as Scrambled Block Hadamard Ensemble (SBHE) measurement matrix and Projections Onto Convex Sets (POCS) reconstruction algorithm are constructed.Comparisons between the proposed underwater imaging method and the existingunderwater imaging methods are also taken by the theoretical calculation on bothimaging distance and image quality. The feasibility of the proposed technology andthe necessity of the proposed frame structure are fully proved by computersimulation experiments. Meanwhile, the superiority of the proposed underwaterimaging technology is fully shown by the theoretical analysis and the simulationexperiments.Owing to the complicated characteristics of underwater optical imagingenvironment, a novel active illumination compressive sensing-based underwaterimaging system using the optical sparsity method is first proposed to increase thesampling efficiency and to decrease the sampling time of the underwatercompressive sensing-based imaging system. By increasing the sparsity of the imagesignal via optical sparsity, it is easy to reach the aim of reducing compressive sensingmeasurements. The system framework of the proposed method is also constructed.According to different light sources, the mathematical models of the proposedsystems using coherent illumination and non-coherent illumination are made up inorder to determine the mathematical relations between the measurements and thetarget signal. In non-coherent illumination case, the constrainedl1-normoptimization algorithm is proposed to overcome the ill-posed equation problemoccurred in the image inverse calculation process. In coherent illumination case, aninverse algorithm with the complex amplitude of light wave field replacing by thespecific value is put forward to solve the problem that the Fourier inverse transformis not implemented because the sampled data in the inverse system do not containphase information. The results obtained from the computer simulation experimentswhen using the non-coherent illumination case as an example shows that theproposed method increases sampling efficiency and improves image quality greatly.In order to overcome the disadvantages of the traditional rage gating, such asthe fixed imaging distance, the complicated hardware and control system and high cost, a new range gating is proposed to reach the goal of increasing imaging distanceand image quality using the principle of range gating and the superiority ofunderwater compressive sensing-based imaging system to reducing the stray light.Meanwhile, the proposed method has many advantages such as flexible imaging,simple system, low cost and less system errors.A new compressive sensing method based on hybrid sampling and blockstrategy (BHCS) is proposed for images to improve the low sampling efficiencyproblem existing in image compressive sensing algorithms available. In the method,a hybrid measurement matrix combining random sampling (RS) and low-resolutionsampling (LRS) is designed to complementally measure the image information datawith high sensing efficiency using the fact that low-resolution sampling canefficiently measure the low-frequency information in image signals. Further, thehybrid measurement matrix with simple structure is proved theoretically to beincoherent with most fixed sparsifying bases. And block strategy ensures that thecomplexity of measurement and reconstruction processes does not change due to theimage size, so the method is simple and easy to implement, suitable for large-scaleapplications. Experimental results show that the proposed method using totalvariation (TV) reconstruction algorithm achieves much better results than manystate-of-the-art algorithms in terms of both PSNR and visual perception，whichmeans a lot to the application of CS theory on the field of real-time processing.