Towards Annotation Free License Plate Recognition

Automated license plate recognition (ALPR) is a key capability in transportation imaging applications including tolling, enforcement, and parking, among others. An important module in ALPR systems is image classification that includes training classifiers for character recognition, commonly employed after detecting a license plate in a license plate image and segmenting out the characters from the localized plate region. A classifier is trained for each character in a one-vs-all fashion using segmented character samples collected from the actual camera capture site, where the collected samples are manually labeled by an operator. The substantial time and effort required for manual annotation of training images can result in excessive operational cost and overhead. In this paper, we propose a new method to minimize manual annotation required for training classifiers in an ALPR system. Instead of collecting training images from the actual camera capture site, our method utilizes either artificially generated synthetic license plate images or character samples acquired by trained ALPR systems already operating in other sites. The performance gap due to differences between training and target domain distributions is minimized using an unsupervised domain adaptation. The efficiency of the proposed method is shown on artificially generated and actual character samples collected from CA and NY.