No abstract

We investigated the potential use of synthetic data for automatic license plate detection and recognition by detecting and clustering each of the characters on the license plates. We used 36 cascading classifiers (26 letters + 10 numbers) as an individual character to detect synthetically generated license plates. We trained our cascade classifier using a Local Binary Pattern (LBP) as the visual descriptor. After detecting all the characters individually, an investigation has been established in identifying and utilizing a clustering algorithm in grouping these characters for valid license plate recognition. Two clustering algorithms have been considered including Hierarchical and K-means. Investigation results revealed that the hierarchical clustering algorithm approach produces better results in clustering the detecting characters than the K-means. Inaccuracy in the actual detection and recognition of license plates is largely attributed to the false detections in some of the 36 classifiers used in the study. To improve the precision in the detection of plate numbers, it is recommended to have a good classifier for each character detection and utilization of a good clustering algorithm. The proponents concluded that detecting and clustering each character was not an effective approach, however the use of synthetic data in training the classifiers shows promising results. Keywords: Cascading Classifiers, Synthetic Data, Local Binary Pattern, License Plate Recognition