Research On Acquisition System Modeling Based Remote Sensing Image Resolution Improvement

Remote sensing images are badly demanded for the military and commercial applications. The development of high resolution and good image quality sensors has become the main focus of earth observation technology. There are three traditional ways to improve spatial resolution: using smaller detector, increasing the camera focal length, reducing the satellite orbital altitude. But due to the inherent limitations of the physical properties, these methods are more and more difficult to achieve new breakthroughs. Image processing provides a new method to improve the remote sensing image resolution. The sub-pixel sampling technology is proved to be effective and feasible; meanwhile, the cost of is less expensive. This paper mainly focuses on the meth-ods of improving the image resolution, especially the effective resolution analysis of sub-pixel technology, image restoration and super resolution reconstruction.(1) The optical remote sensing image system is studied. The image acquisition system includes the atmosphere, platform movement, optical system, detectors and noise. We analyzed the limits of the sub-pixel technology from the perspective of sampling theorem. Using sub-pixel technique can take advantage of the optical system. If the detector size and the optical system aperture are decided, the sub-pixel technology can only infinitely approximate to the system optical diffraction resolution but could not go beyond this limit.(2) A novel single image super-resolution reconstruction method with frequency domain correction is proposed to solve the ill-posed problem imported by zero-crossing-points of mod-ulation transfer function (MTF). The satellite movement and some other factors bring lots of zero-crossing-points below the cut-off frequency. Noise near these zero-crossing-points would be amplified in the image restoration process, even cause false edges. To solve this problem, the geometrical characteristics of acquisition system are considered in the frequency domain cor-rection algorithm. In our method, the total variation restoration is used to revise the frequency spectral above the cut-off frequency; the wiener filter is used to initialize the TV restoration;the frequency domain correction can reduce the artifact caused by zero-crossing-points. Experi-mental results indicate that the proposed methods considering the modeling of image acqui-sition systems reconstruct high-quality super-resolution image for both composite image and remote sensing images with different physiognomy.(3) Remote sensing image is degraded by a variety of factors. Aliasing is one of the important factors which affect the image quality of optical sampling remote sensing imaging system. After analyzing the aliasing-blurring-noise model in push-broom linear CCD imaging system, we quantitative calculated the aliasing, blurring and noise in sub-pixel sampling system with effective resolution. The tilting sampling mode includes the single linear tilting sampling and the super-mode tilting sampling. These two modes get a better resolution by rotating the CCD linear with a degree θ and/or reducing agent the sampling distance d. The impact of different parameters is discussed by calculating the effective resolution and optimal parameters are suggested.(4) A novel image restoration method in the MAP-MRF frame is proposed to reduce the impact of aliasing and blurring. The optimal reciprocal cell is designed based on the effective resolution calculation. This optimal reciprocal cell is used to maximize the tilting sampling remote sensing image information. In the frame of MAP-MRF image restoration method, the optimal reciprocal cell is coupling with Huber-MRF image prior to reduce the blurring, aliasing and noise. After the restoration, the higher frequency part dislocated by sampling is corrected. A majorization-minimization approach is used to reduce the computational complexity. Exper-imental results indicate that the proposed method considers the modeling of image acquisition systems and obtains an effectives image restoration result, improves the image effective resolu-tion.(5) Super-tilting mode improves the spatial resolution by rotating the two dedicated linear arrays with a degree. The bias-correct and interpolation is need to get uniformity image. The framework of super-tilting mode image super-resolution is built with registration, interpolation and deblurring. We use Forstner operator to extract image futures. Correlation is used to get a rough match, and then a distance constraint is used to remove the wrong matches. After the rigid registration, the angle of linear arrays is added into the transformation model, and a B-spline based interpolation is used to get the initial image of the high-resolution. Then we use the MAP based super-resolution reconstruction with optimal reciprocal cell to get the final result. Experimental results indicate that the proposed method reduces the aliasing in recovery and obtains an effectives image super-resolution restoration result.