Research On The Super-resolution Technologies Of Digital Images

After the theoretic feasibility of super-resolution reconstruction proved by Harris, the researches on image super-resolution reconstruction were not interrupted as time passed. At first, the researches on image super-resolution reconstruction went along with the instruction of Fourior Optic Thoery. With the increasing of the application fields of super-resolution reconstruction, however, a lot of research methods of super-resolution reconstruction appeared. These methods include the super-resolution reconstruction from single image to single image, from mutiple images to single image and from image sequences to image sequences. The typical example of the development is to gain a high- resolution image rather from the mutiple remote sensed images than from the single remote sensed image. And their application areas are enlarged from remote sensed image reconstruction to other sorts of image reconstruction, such as super-resolution reconstruction of degraded images in monitor systems and low-resolution image sequences through low capacity channel. In spite of so wide developing perspect, the researches on image super-resolution reconstruction are still greatly immature, and there are plenty of unsolved works on the design of its algorithms and development of its theory. In this thesis, three aspects about super-resolution reconstruction will be developed, including the generalization of POCS algorithm, an improved model of imaging process and build-up of estimation criteria of super-resolution reconstruction.The first part of the thesis introduces the development of super-resolution technologies briefly, points out their military and public application fields, summarizes the main methods of super-resolution technologies in detail, and expatiates the difficulties and future of the technologies.In the following work on super-resolution algorithms, a generalized POCS algorithm is developed on the basis of the traditional POCS algorithm. The generalized POCS algorithm introduces the statistics of the noise to the constraints of projection iteration so that the algorithm is able to eliminate the noise effectively. Nowadays, the major algorithms in super-resolution reconstruction are done on the linear space-invariant image restoration. But in real-world applications, the degradations are often space-varing and the degrading system PSF changes with the spatial location over the image. And the generalized POCS algorithm in this thesis is mainly designed for the space-variant degrading PSF, which alters with the different spatial location over the image. Through some typical experiments, the generalized POCS algorithm exceeds the Wiener filters in removing space-variant blurs and surpasses Bi-linear algorithm in super-resolution reconstruction from mutiple images to single image.On the research of imaging theory, the thesis points out the deficiency of Harris theory, novelly introduces a window process and developes an improved model of the digital image formation. Based on this model, the aliasing in the frequency domain of digital images is pointed as another important factor decreasing the resolution of digital images besides Optic Limited Diffraction Frequency by analysing the frequency traits of various signals in the digital image formation. Furthermore, we define the resolution of ditigal images tentatively on the basis of the aliasing in the frequency domain. In the experiments of this part, we creatively compare the Discrete Fourior Transform coefficients of variant-resolution images of the same scene to prove the existence of the aliasing in the frequency domain of digital images.In the last part of the thesis, the estimation criteria of super-resolution reconstruction are noticed, which were rarely mentioned in other papers before. Based on the Discrete Fourior Transform coefficients of variant-resolution images of the same scene, a called Frequency Residual is developed to compare these discrete fourior coefficients. For the new estimation criterion, we design a contrast experiment on the generalize...