Symbol occurrence probability vectors for syntax correction in automatic number plate recognition systems

摘要

Automatic number plate recognition systems (ANPR) are starting to be used in surveillance and access monitoring applications. They are mostly based on the detection of the number plate, the segmentation of characters and the optical character recognition. In most of the cases, because of the higher degree of generalization, the syntax is not an a priori element of those systems, therefore reducing the recognition rate. This paper presents new features used to improve the performances of ANPR system by using the country standard information in order to determine character/digit occurrences for each symbol position in the number plate. The proposed ANPR system is composed of three individual classifiers which perform different. In order to exploit their individual performance, trainable and non-trainable decision fusion rules are used. The syntax correction is implemented in the final stage of the system, as a feature vector for the decision fusion meta-classifier.