A License Plate Super-resolution Algorithm based on Dictionary Learning and Neural Network
Xin Zhang, Xiao Liu, Zhu Qiang Xiao, Li Ming · 2021
A license plate super-resolution algorithm was proposed based on dictionary learning and neural network. Firstly, a high-resolution dictionary was built for each number on license plate. Secondly, the dictionary was applied to each blurry number image respectively to fill in the missing feature information. Then, the images with features filled were input into the neural network to perform the feature extraction, pixel shuffle and other operations. After that, the final super-resolved image was obtained. Experimental results showed that proposed algorithm outperformed existing super-resolution algorithms like Bicubic, SRCNN, SRGAN and VDSR as applied to blurry license plate images, and greatly improved the recognition rate.