Super-Resolution Towards License Plate Recognition

Abstract

Recent years have seen significant developments in license plate recognition through the integration of deep learning techniques and the increasing availability of training data. Nevertheless, reconstructing license plates from low-resolution surveillance footage remains a challenge. To address this issue, we propose an attention-based super-resolution approach that incorporates sub-pixel convolution layers and an Optical Character Recognition (OCR)-based loss function. We trained the proposed architecture using synthetic images created by applying heavy Gaussian noise followed by bicubic downsampling to high-resolution license plate images. Our results show that the proposed approach for reconstructing these low-resolution images substantially outperforms existing methods in both quantitative and qualitative measures. Our source code is publicly available at https://github.com/valfride/lpr-rsr-ext/.