Licence plate recognition presented in this paper is used to identify vehicles by their licence plate numbers. This technology can be widely used for paying pay-rolls, in opening parking garage door, traffic control /1/, etc. This paper presents an algorithm for licence plate recognition, shown in Fig. 1. The algorithm itself is divided into two parts, the first part extracts licence plate and the optical character recognition is described in the second part of the paper. In the first part of the paper three methods for licence plate recognition are presented. White pixels in an image can be detected using threshold method. Reference photo is shown in Fig. 2; meanwhile the experimental result of threshold method is shown in Fig. 3. The second method is based on 2D correlation, which uses segmentation in order to limit search area. The segmentation technique is performed using (1), but it is computationally very demanding (2). The result of this phase is shown in Fig 4. The method with Euclidian norm (3) uses Freeman chain code with interior pixels, but it is user dependent, because the user has to input one licence plate pixel (the result is shown in Fig. 5). The first method is used as method for licence plate recognition. The dilatation algorithm is used to fill up the whole plate with the same pixels. The dilated picture is shown in Fig 6. The algorithm detects the angle of licence plate, as shown in Fig. 7. Rotated and isolated licence plate is quantised and rescaled using a histogram method. The extracted licence plate is further shown in Fig. 8. The edges cut from detected licence plate and extracted binary images are shown in Figs. 9-10, respectively. A method called peak-to-valley is used to extract each individual character, which sums picture's columns and creates a histogram. By comparing sums to specified threshold, characters are detected. Characters are recognised using feed-forward neural networks. This network has 200 input neurons and 36 output neurons. Learning is accomplished through supervised learning of back-propagation technique. The whole structure of neural network is shown in Fig. 11. The results presented in Table 1 show that the overall efficiency of OCR engine is 96% when the recognition is applied on extracted licence plate
