Attributes Selection for Licence Plate Recognition Based on Decision Trees

The paper deals with an image-based recognition system, which captures, interprets, records, and processes the image of a license plate for use in a variety of ITS applications. Such a system can save money by collecting and processing vehicle data without human intervention. Motorists are then allowed to pass toll plazas or weigh stations without stopping, which can save their time and prevent occurrence of congestions. Helping control access to secured areas or assisting in law enforcement can also improve safety and security. In different references this technology is also referred as Automatic Vehicle Identification, Car Plate Recognition, Automatic Number Plate Recognition, Car Plate Reader or Optical Character Recognition for Cars. The paper discusses a concept of the Licence-Plate Recognition system consisting of several modules that are in different stages of development. It focuses on processing of images with only one vehicle captured. The system is based on decision trees created using inductive tree algorithms. After brief outline of the system the main attention is paid to selection and detailed description of attributes used to build a decision tree. The paper includes an algorithm used to find holes and arcs in the characters. Totally 32 principal attribute types are explained. Some of them are used several time in an analogical way, thus the final decision tree of the discussed system prototype contains 77 nodes. The system has been prototyped using C++ and designed for recognition of Slovak-style license plates. Some of introductory procedures (image input and selection of area with licence plate in the image) are still performed manually and in future can be fully automated.