In this paper, a new algorithm for the identification of a vehicle license plate is proposed, on the basis of feature analysis image processing joint with Artificial Neural Networks. More specifically, the ANN was trained to identify alphanumeric characters from car license plates based on data obtained from algorithmic image processing. For this reason, the system was tested with 40 natural scene images of 500x424 pixels, using a common digital camera. The screen shots were taken from various distances. The algorithm includes enhancement, geometric operations and morphological features identification. In all 40 images the license plate was properly segmented (success 100 %). The present ANN is a two layer Probabilistic Neural Network with biases and Radial Basis Neurons in the first layer and Competitive Neurons in the second one. Its performance reached 92.5 %, which is completely correct character identification in 37 of 40 plates. In the remaining 3, there was an identification error in one of the characters. The test images have different backgrounds such as other vehicles and buildings under different illumination conditions. The variety of these conditions demonstrates the robustness of the algorithm, since no restrictions were imposed in the input images. A future development is the implementation of this algorithm in hardware, which would reduce the processing time.
Digital image processing and neural networks for vehicle license plate identification
Die digitale Bildverarbeitung für die Erkennung von Kraftfahrzeugkennzeichen mit neuronalen Netzwerken
2000
6 Seiten, 4 Bilder, 1 Tabelle, 12 Quellen
Conference paper
English
European Patent Office | 2022
|