Super-Resolution of License Plate Images via Character-Based Perceptual Loss

摘要

License Plate Recognition (LPR) is an highly influential problem in computer vision. In this paper, we present a super-resolution model specialized for the license plate images, CSRGAN, trained with a novel character-based perceptual loss. Specifically, we focus on the character-level recognizability of the super-resolved images rather than the pixel-level reconstruction. Experimental results validate the benefits of our proposed method in both quantitative and qualitative aspects. In particular, our method achieves a higher character-level recognition accuracy over the state-of-the-art image super-resolution techniques.