Research On Superresolution Of License Plates In Surveillance Video

Super resolution technology can be used for license plates images reconstruction to enhance the image quality,which can produces a high resolution plate image from a set of low resolution images. Nowadays, research on super resolution construction has become a hot topic in the field of image enhancement.We focus on the Bayesian superresolution of license plates image with affine transformation.Image registration only contains estimate of translation and rotation in traditional superresolution algorithm,which can not accurately describe the movement of the license plate in the surveillance video.By introducing affine transformation to the registration, the license plate movement get a more accurate representation. To find the distribution on the HR image, a unique approximation is obtained by minimizing a linear convex combination of Kullback-Leibler(KL) divergences. The distribution of unknowns are transformed into tractable ones by using Taylor expansion approaches. We find this distribution satisfies the form of multivariate Gaussian distribution through calculation. The bayesian framework is adaptive to the change of the model parameters in iteration process without supervision. Finally, we get the reconstructed image and affine registration parameters.The estimated HR images are compared with the ones obtained by other SR reconstruction methods. Experimental results show that this algorithm can effectively produce a high quality plate image from a set of low resolution plate images, which we can effectively identify the plate characters.