Deep Super-resolution Method via Generative Adversarial Networks for License Place Image Enhancement

Abstract

Convolutional neural networks and generative adversarial neural networks have recently shown outstanding performance in single-image super-resolution. In this paper, we propose a deep super-resolution method based on generative adversarial networks to reconstruct a high-resolution license plate image from a low-resolution license plate image to improve the accuracy of lowresolution license plate recognition. To achieve this, we create a super-resolution model using deep residual blocks. In addition, our model uses the benefits of adversarial loss, pixel-wise loss, and perceptual loss. Adversarial loss plays a role in enabling the discriminator network to distinguish between the high-resolution image and the fake high-resolution image generated by the generator network well, while it facilitates the generator network to generate a realistic high-resolution image with the goal of fooling the discriminator network. Pixel-wise loss and perceptual loss help the generator network to reconstruct a better high-resolution image. We compare the proposed method and other state-of-the-art methods with the performance of the license plate image recognizer. The experimental results demonstrate that our proposed method performs well by showing that the accuracy of the lowresolutionimage is improved from 38.45 to 73.65% by the proposed method.