Research On Blind Restoration Of Remote Sensing Images Based On Huber-markov Random Field

During the imaging process(achievement, transmission and storage), satellite-based remote sensing images are subject to various blurring phenomena like atmospheric turbulence, scattering, spacecraft motion, diffraction of the optical system, defocus, distortion and other properties of the sensor and atmosphere. Images we obtain are always blurred and contaminated by noise. Image restoration technology is able to get the restored result which is closed to the original image from the degraded one. Point Spread Function(PSF) of the imaging system is the key of restoration but it is hard to obtain PSF precisely as the complicated imaging process. Thus, ordinary restoration theories are struggled to restore remote sensing images. Blind image restoration focuses on the essence of degradation which is not enslaved to different degradation processes and it is able to estimate PSF only by the analysis of the achieved image without the knowledge of degradation process es than restore the data of which the outcome will be more precise. It is a really good solution to the absence of prior knowledge.This thesis focuses on the study of blind restoration for optical remote sensing images based on the Huber-Markov random field(HMRF). In order to obtain good edge information and protect texture information, HMRF model has been concluded into consideration for its limited solution space and quick access to the best solution. As the blind restoration problem is about getting PSF and doing inverse operation and the change of Point Spread Function(PSF) is the essence of degradation, it is vital to estimate PSF exactly. This essay estimates the PSF based on knife-edge method and optimize it based on HMRF model for the consideration of imaging characteristics of optical remote sensing images and the fact of absence of prior information. The method proposed here supplies an applicable, stable and accurate scheme for the blind restoration of optical remote sensing images and it is practically valuable.This thesis introduces the development of this domain in details ranging from blind restoration theory, Markov random field theory and edge extraction technology. Then it introduces details about all the theories above and proposes a new optimized PSF estimation method based on knife-edge method and HMRF theory. These two theories make it come true for the exact extraction of edge information, the construction of two-dimensional PSF and optimization of the PSF in further. At the meantime, this essay elaborates the process of this method including the construction of the object function combining probability statistics theory for building image model and the iteration process to finish the blind restoration task. Considering the huge data, we propose a n iteration optimization method based on the sparse representation theory to solve the data problem. In the end, this essay uses a new evaluation method where Modulation Transfer Function(MTF) and Full-Width at Half-Maximum(FWHM) are used as detail evaluation criterion at the same time other non-reference criteria are used to evaluate the whole restored image. This new method supplies a new perspective to evaluate restored image from not only wholesome information but also details of the image. Results show that this proposed method has clear outcome and it is applicable.