Research On Single-image Super-resolution Reconstruction

In practice, because of the limitation of imaging equipment internal hardware we often don’t have access to high resolution image. Since the cost to improve the resolution of the image by improving the quality of hardware is quite high, and some technical problems involved are difficult to solve in the short term, so it has significant meaning to improve the resolution of the image from the aspects of software and algorithm. The process to obtain high resolution image that meets the requirements from a low resolution image or sequence by signal processing technology is referred to as image super resolution reconstruction.Because digital signal processing technology is mature and perfect and it needs not to change the existing equipment, image super resolution reconstruction technology has significant advantages both in technology and cost, so it has been widely used in the fields of public security, medical imaging and high-definition digital TV and military remote sensing monitoring and so on. Relative to multi-frame super resolution reconstruction technique requires more low resolution image sequence under the same scene, single-frame image super resolution technology only utilizes one low-resolution image of the actual scene to estimate the required high-resolution image of the same scene in image reconstruction, which can meet more the demand in practical application. Therefore, this paper mainly studies single image super resolution reconstruction technology.In this thesis, firstly several classic image super resolution reconstruction algorithms at present such as interpolation algorithm and kernel ridge regression algorithm are described, and their performance characteristics are introduced analyzed, then the experimental simulation and comparative analysis are given. Known from the analysis of simulation results and discussion the interpolation based algorithms are simple, but the reconstruction image smooths the details seriously and then results in serrate and blurred edge. It is suitable for the image demanding higher real-time performance but with fewer details. Kernel ridge regression algorithm for image super resolution reconstruction has better reconstruction effect on images with clear main edge, but has poorer performance when the image contains many tiny edges, and takes longer time. Therefore, in this paper, a new single image super resolution reconstruction algorithm based on the bidimensional empirical mode decomposition is put forward.The bidimensional empirical mode decomposition can decompose the image into several layers of different frequency adaptively, the first layer contains the main edge detail information, the detail information of the following several layers reduces gradually, and the final residual contains only the information of brightness and the general trend. Employing the advantage, the first layer containing the main edge detail information is reconstructed with kernel ridge regression algorithm in order to ensure the reconstruction image quality; the following several layers are reconstructed with interpolation algorithm to reduce the operation time. A large number of simulation experiments about the algorithm in this paper are made to compare and analysis with several other algorithms.Known from the simulation results and analysis, the reconstruction effect and computing speed are both best when the image is decomposed into 4 layers by bidimensional empirical mode decomposition. The experiment and comparison prove that the algorithm in this paper sufficiently combines the advantages of kernel ridge regression algorithm, bicubic interpolation and bidimensional empirical mode decomposition, which both guarantee the quality of reconstruction image and reduces the computing time, and verifies the superiority and effectiveness of the algorithm in this paper. At the same time, compared with the learning algorithm based on neighbor and sparse algorithm, the algorithm in this paper is a kind of effective, superior image reconstruction algorithm from the visual effect, objective indexes and computing time.