| License plate (LP) recognition is an important part of intelligent transportation system (ITS).License plate image enhancement is the first step of license plate recognition system. When the LPR system is employed to identify vehicles, the captured image resolution tend to be low due to some objective factors. Therefore, in order to improve the license plate spatial resolution, the super-resolution reconstruction (SR) technology is applied to enhance the plate image.Combined with the characteristics of our license plates, in this paper we formulate a model for super-resolution which is based on total variational regularization. The algorithm includes three steps:The first step, motion estimation is based on TV-L1optical flow: one-dimensional optical flow and multi-dimensional optical flow.The model uses the total variation regularization algorithm and efficient method-by-point threshold. Reduce the comp-utational complexity. The second step, we have formulated a new time dependent convolutional model. We have achieved the total variation regularization super-resolution reconstruction algorithm. The third step, we use the Bregman iterative and inverse scale space method to improve the spatial resolution.Experimental results show that the reconstruction algorithm has strong robustness, and this Algorithm has great advantages in property and computing speeding. |
PDF Full Text Request