Research For Super-resolution Reconstruction Technologies Of Image Based On Airborne Photo-electricity Imaging System

Image acquisition process is easily affected by optical blur, motion blur, aliasing,noise etc., which reduce the quality of images. However photo-electricity equipmentof aviation need to obtain images or videos as clear as possible without increasing itscost and size unboundedly, that means we should not change the hardware structure ofphoto-electricity device. Therefore, the software method of high-resolution (HR)images or videos reconstruction by using the complementary information of asequence of low-resolution (LR) images becomes a necessity.Because of the merits of DSP chip such as highly integrated, stable, fast speed,and so on, this paper has proposed to realize super-resolution reconstruction on theembedded image processing platform which core is TI’s DM642. So it improved theimaging system resolution greatly without changing the structure of system.Firstly, this article introduced a degradation model of imaging system, and foundthat the main factors of impacting the quality of images of super-resolutionreconstruction included the estimation accuracy of motion parameter of imagesequences, the number of images, the estimation accuracy of point spread function(PSF) of system, the elimination and suppression of noise, the radiance differencesbetween images or sub-blocks and the algorithm of super-resolution reconstructionetc., and then explained several key parts of these factors in detail.Secondly, this paper has expounded and analyzed the motion estimationalgorithms based on logarithm polar Fourier transform and three-parameter Taylorexpanding. It has raised a motion estimation algorithm which firstly extracted theedge of images adaptively, and then used the improved Fourier Mellin transformalgorithm for coarse estimate, lastly realized accurate motion estimate with three parameters Taylor expansion algorithm based the affect of the accurate of motionestimation of LR image sequences to super resolution. The accuracy of improvedmotion estimation algorithm is0.05units, and the time of estimation reduced to1sfrom3.5s for motion estimation of size256×256image, and can realize accuarateestimation for images which’s detail is not obvious.Thirdly, it has proposed a pre-processing method which carried through imagedenoising and then adaptively removed bad images which were still not yet the idealimages, so it could prevent the noisy image and the degraded seriously image fromaffecting the quality of reconstruction. Experiments showed that the pretreatmentmethod of noisy or blurred images greatly improves the quality of imagereconstruction.Fourthly, this paper did a research on the reconstruction algorithm based onPOCS in-depth and analyzed the merits and demerits of this method, then found thealgorithm is faster than other algorithms but does not have strong ability of noiseimmunity. Based on the considerations of speed and the noise immunity, this paperpresented a method which carried out pre-processing of image sequences, and thenusing the improved POCS method to reconstruction. Improved reconstructionalgorithm reduced the reconstruction time from4.3s to1.5s for6frames images of128×128image sequence (reconstruction factor is two), and improved the noiseimmunity.Finally, it has analyzed the key factors of impacting the effectiveness of thealgorithms of motion estimation and reconstruction, and then cut and improved thealgorithms, and ported the algorithms to the embedded image processing platformwith the core of DM642DSPs, optimized some key code by linear assembly, greatlyreduced the image reconstruction time. The resolution of camera is720×576. thereconstruction time takes about20s. Although the time of reconstruction is long, andcan not realize real-time, but it has been ported to the hardware platform and lay theinitial foundation for future.