Skip to Main Content
This paper proposes a soft recognition-guided super-resolution (SR) approach to high-resolution (HR) vehicle license plate (VLP) reconstruction. In license plate SR, conventional algorithms often do not use prior information of alphanumeric to reconstruct the HR image. There are several existing learning-based methods that employ HR example patches as prior to reconstruct the HR image. However, they utilize a hard binary decision strategy to determine whether the local patches are fully relevant or totally irrelevant. This may be inconsistent with many practical applications as the degree of relevance for example patches may vary. In view of this, this paper proposes an iterative SR method that incorporates soft local patches learned/reinforced from optical character recognition (OCR) of the reconstructed VLP into HR license plate reconstruction. Experimental results show that the proposed method is effective in performing license plate SR.