Learning Based Character Segmentation Method for Various License Plates

Abstract

This paper proposes a novel method for license plate character segmentation using a classifier. Conventionally, a three-step method of detection, segmentation, and recognition is commonly used for license plate character recognition systems. Although a machine-learning based method is widely used in the detection and recognition steps, only a heuristic method based on a projection or connected component analysis is used in the segmentation step. The method proposed in this paper, however, uses a machine-learning based method for segmentation, unlike previous researches. The proposed method consists of several steps. First, locating the common region in the license plate, extracting the area containing all license plate types from this region, classifying it to determine the license plate type and estimating the location of the remaining characters based on the structural information of the determined license plate type. Experimental results show that the performance rate of the proposed method is 98.2% for over 10,000 license plate images.