Single-point Imaging System Based On Compression Sensing And Its Applications

Nyquist sampling theorem points out that sampling frequency of band-limited signal must not be less than twice of the highest frequency in signals in order to reconstruct the original signal. However, the hardware cost will increase if the signal bandwidth and therefore the sampling rate increase. It was becoming increasing difficult for the corresponding hardware equipments to deal with such broadband signals.Compressed Sensing (Compressive Sensing, CS) is a new image processing theory. It offers a new way to solve those problems. Different from sampling theorem, CS suggests that as long as signals can be sparse representation by some kinds of transformation; global sampling of the sparse signal will be implemented. That is CS could represent compressible signals at a sampling rate significantly below the Nyquist frequency; then few global measurements can be got and the original signal can be accurately reconstructed from those measurements by certain reconstruction algorithms. It breaks through the bottleneck of Nyquist sampling theorem and makes it possible to deal with broadband signals.Currently the main researches on compressed sensing mostly were focused on theoretical research and simulation of reconstruction algorithms. Researches rarely involved in designing a hardware system to apply CS into practice. The main reason was that it was very difficult to implement the measurement matrix by some kinds of hardware. Moreover, it lacked properly practical reconstruction algorithms.Compressed sensing was discussed in detail from sparse representation of signals, measuring encoding and reconstruction algorithms. Three reconstruction algorithms, including, matching pursuit algorithm, orthogonal matching pursuit algorithm and regularized orthogonal matching pursuit algorithm, were presented. The minimum mean square error of the linear estimate (MMSE) algorithm used in the developed system was also investigated. Through the comparison of image simulation reconstructions using those algorithms, MMSE showed better reconstruction quality under low sampling rates and great superiority. It demonstrated that MMSE had great potential in applications.A manual single-point imaging system was investigated in this thesis. A series of optimization masks were taken as the measurement matrix and MMSE was used to reconstruct several characters samples in experiments. Compared with the simulation results, it was demonstrated that the imaging system had good performance. However, the manual imaging system had low efficiency and relatively poor flexibility. Therefore, an automatic rotating disk imaging system was developed based on CS algorithm. In this system, the rotating disk was controlled by a motor and a series of measurement matrix was got according to the number of measurements. This system was more efficient and had better flexibility than the manual one. Character samples were reconstructed with different numbers of measurements using the automatic system. The stability and reliability of the automatic rotating disk imaging system were demonstrated.Except the simulation of image reconstruction, manual and automatic single-point imaging systems based on CS algorithm were investigated and developed in this thesis. Character samples were reconstructed with high quality under low sampling rate by using the developed system. It was demonstrated that the imaging system was reliable with good performance and had great value and potential to use in applications.