Research On Aerial Imaging Motion Deblurring Technology

In the process of aerial imaging, the camera moves at a high relative speed to the ground, which inevitably brings image motion on the CCD focal plane for the target scene and results motion blurred images. To improve the aerial imaging resolution and quality, the image motion must be compensated. In this dissertation, the compensation of image motion is done by image restoration using image motion deblurring methods.At first, several image motion compensation techniques are described which elicits the research content of the paper. Then the research process of image motion deblurring technology is presented. Based on the imaging and object motion model, we get the image motion blurring model. The deblurring model in this paper is based on the maximization a posteriori (MAP) framework, and both the motion blur kernel and the original image are iteratively optimized with an alternative minimization (AM) method simultaneously.With respect to the problem of motion deblurring from a single blurred image, the paper proposes a method based on constraints of local region of the blurred image. Using different priors of the local region and the motion blur kernel, we get a minimization energy function which is solved by the AM method, and then we estimate the motion blur kernel. Based on the blur kernel estimated, the entire original image is restored by Richardson-Lucy deconvolution.The motion blur kernel can be depicted by the motion path of the camera during the CCD exposure time and more information can be utilized from multiple images which can reduce the ill-posedness of the motion deblurring problem. Considering these matters, we proposed a motion deblurring method based on multiple images. We use a high frame-rate but low resolution camera to record the motion path of the low frame-rate but high resolution imaging camera during its exposure time. By taking advantage of the optical flow of the multiple low resolution images, we estimate an initial blur kernel, and then we combines both multiple low resolution images back-projection and multiple image deconvolution to refine the motion blur kernel and the clear image with the AM method. The original image is estimated in the end. Using our method, we can estimate the blur kernel at a rather correct accuracy and further improve the deblurred image quality.In regard to the problem of image deblurring in the presence of Gaussian and salt-pepper noise, we first detecte the salt-pepper noise points and exclude them in the deblurring process, and then we adopte a refined variation based method to restore the original image. Using this method, the salt-pepper noise is restrained effectively and the quality of the restored image is improved.