Research On Super-resolution Reconstruction Algorithm Based On L1Norm

In the detection of road diseases, images used to further analysis are usually lowresolution images. Besides owing to the technical difficulties of imaging system and thehigh cost, higher resolution images are difficult to obtain. The fundamental of thesuper-resolution reconstruction is to reconstruct higher resolution images from asequence of low resolution images with complementary non-redundant information.Image super-resolution reconstruction improve resolution in software, so it doesn’thas to change the hardware of the imaging system, therefore it can greatly reduce thecost. There still exist many problems to be resolved in the field of super-resolutionreconstruction. For example it is an ill-posed inverse problem and doesn’t have a uniquesolution. And the whole process of super-resolution reconstruction involves many links,such as motion estimation, interpolation, de-blurring and de-nosing, so it is important tocoordinate all parts to find an efficient way.The main research contents and contribution are as follows:①An improved small cross-diamond search algorithm based on GaussianPyramid.In order to overcome the high complexity of the motion estimation, a newdiamond search algorithm based on Gaussian Pyramid is proposed.In this case, initialmotion vector is estimated in the upper layer of the Gaussian Pyramid, and the finaloffset value is based on the initial motion vector produced in the upper layer.Experimental results show the proposed method could speed of128.24%,96.55%and16.82%over the DS, NDS and AHSDS with the small reduce in search accuracy.②An iterative blind deconvolution based on forecast weight: in this paper, we willuse an iterative blind deconvolution based on forecast weight and in the process ifiteration, the acceleration parameter will be updated at the end of every iterative stepaccording weights. Experimental results show that the proposed method could speed of48.2,51.43,50.43,43.9times over the L-R algorithm with the small reduce in searchaccuracy.③In a word, we can conclude that the proposed small cross-diamond searchalgorithm based on Gaussian Pyramid and the iterative blind deconvolution based onforecast weight could improve the efficiency.And the algorithm based on L1normworks well in maintaining the edge of the image. Experimental results shows that theeffect is significantly improved...