Real-time Restoration Algorithms Research For Aerial Images With Different Rates Of Image Motion

When reconnaissance plane is flying with high-speed and low-altitude, the relative motion will happen between the airplane and the ground object. During the reconnaissance camera's CCD integration time, image motion will occur on the focal plane array of the aerial camera to blur the images and degrade the reconnaissance quality. The flight attitude's difference will lead to different types of image motion. If the image algorithm is used to restore all kinds of motion blurred aerial images for the convenience, the validity and the efficiency of the presented restoration algorithm will be strictly requested.At the first, this article studies the simple forward motion. A new algorithm based on one-dimensional wiener filter (1DWF) is proposed to restore the two-dimensional images blurred by forward motion. The new algorithm guarantees the restoration quality and reduces the algorithm computational requirements to original 1/3. Based on this, the mathematic model of the different rates image motion generated by aerial side-oblique imaging is founded According to their different image motion rates, the entire blurred image was segmented into many slices. 1D Wiener filters are used in parallel to restore all the image slices to realize the side-oblique image motion blur quick recovery.The rotational motion blur is more complicate than side-oblique image motion blur. An algorithm is proposed to restore the images in real-time, blurred by rotational motion during aerial imaging. In the algorithm, the coordinate conversion based on Bresenham algorithm is applied to transform the space-variant blur on the circular arc paths into space-invariant blur on the pixel lines, and the 1D Wiener filter is used to restore the linearity image motion transformed by coordinate conversion.In the sole image motion blur research foundation, an algorithm is proposed to restore the images, simultaneously degraded by multiple blur during aerial imaging in advanced. To reduce the accumulating of calculation error during the restoration for multiple blurred images, the point spread functions (PSFs) of each space-invariant blur in the multiple blurred images are combined by convolving all the linear PSFs together. For that, the multiple blur made up of the space-invariant blurs can be eliminated by a deconvolution restoration.While realizes effective restoration to each kind of image motion blur, the computation consumption reduction of each algorithm is carried on. The improved real FFT algorithm is applied to instead the conventional complex FFT algorithm used in the original restoration algorithm to reduce the computation consumption by about 50%. Based on this improvement, an algorithm is proposed to restore the motion-blurred color images in real-time by combined the color components, which makes proposed algorithm 3X faster than the general algorithm.In the foundation of algorithm's efficiency optimization, the GPGPU technology is applied to transplant the algorithm to the graphics processor unit(GPU) platform, Using the parallel computing ability of GPU's special simultaneous operating mode, the algorithm operating speed is further promoted. The proposed algorithm can restore a 1024x1024 24-bit motion-blurred color image in 7.95ms on GPU, and the PSNR of the restored image is 31.45. In the end, an overall project concerning all real-time restoration algorithms proposed in this paper is presented...