License Plate Image Super-Resolution Based on Convolutional Neural Network

To improve the visual quality of the low-resolution license plate image in the video surveillance system, this paper proposes a new method of super-resolution, namely multi-scale super-resolution convolutional neural network (MSRCNN), it was inspired by Inception architecture of GoogLeNet. The proposed method uses different sizes of filters for parallel convolution to obtain different features of the low-resolution license plate image, then the features can be fused by the layer of concatenation. Finally, the high-resolution images can be reconstructed through non-linear mapping. Experimental results prove that the proposed method can improve the quality of low-resolution license plate image obviously.