Based On Compression Perception Of Single Image Super-resolution Reconstruction Algorithm Study

Super-resolution (SR) image reconstruction refers to a technology that improves image resolution by using digital signal processing methods, with the ability of still utilizing existing optical imaging systems. This technology has huge commercial values because of its low cost and good effect. It has been widely used in security surveillance, medical imaging, biometric identification, military and civilian remote sensing, etc.Compressive sensing (CS) theory asserts that one can capture and represent compressible signals at a rate significantly below the Nyquist rate. Moreover, the signals can be correctly reconstructed from these projections using an optimization process. The theory provides a new approach for signal acquisition, image compression, the ill-posed problems inverse and other issues. This thesis builds a new SR image resolution model based on CS theory, dealing with the single image scale-up problem, and gets a pleasing result.The article first introduces the concept of SR image reconstruction and the status quo, and then provides an overview of the existing SR technology, mainly learning-based single image SR methods. CS relies on two principles:sparsity and incoherence. This thesis focuses on signal sparse representation, design of measurement matrix and reconstruction algorithms. On this basis, a common SR image reconstruction framework based on CS is proposed and demonstrated. Considering wavelet basis or redundant dictionary as sparse basis, two kinds of single image SR reconstruction algorithms are designed. Finally, these algorithms are simulated and compared with other SR methods. Experiments show that the common framework is a good method to introduce CS theory into the fields of SR image reconstruction. Moreover, the SR algorithm based on compressed sensing can achieve good reconstruction results.