Research on the Optimization of License Plate Character Segmentation

Abstract

License plate character segmentation is an important step in license plate detection and recognition. With the development of machine learning technology, the segmentation algorithm of license plate characters based on clustering is also developed rapidly. However, the current clustering algorithm based on K-means does not consider the integrity of the characters and the horizontal difference between the characters on the license plate. This paper presents a weighted distance measurement method based on a connected domain. First, all pixels belonging to the same connected domain are naturally clustered into a class. Meanwhile, the horizontal and vertical distance measurements between the connected domains are scaled by the respective land weights, which makes the horizontal distance between the connected domains get more attention. This paper conducts extensive experiments on the collected data set of license plate characters, and the experimental results verify that compared with the K-means clustering algorithm, the weighted distance measurement method based on the connected domain has more accurate license plate character segmentation results.