In the previous research on license plate extraction, various techniques have been suggested: a method which employs the Hough conversion of binary image data and extracts vertical line and horizontal line; a method which extracts number regions with predetermined range, based on the differences in brightness and then extends the regions; a method which employs Hough conversion for the regions with high difference in brightness; and a method which digitizes through histogram based normalization and then extracts the number regions. The traditional methods, even though they claimed good performance in terms of computation time and accuracy, suffer from long processing time and lower recognition accuracy than expected in the case of different size of the plates and noisy data from the plates. In this research a novel method is suggested which utilizes the differential value of symmetric brightness (brightness vector) to figure out the possible region of license plates and then employs edge-extraction on partial image to extract the image of the license plate and verification which checks some wrong extractions. This approach makes real-time extraction possible, regardless of the noise, and regardless of the license plate's size. In experiments, it was found that the extraction time of the algorithm for 72 images was 0.27 through 0.55 second, and the accuracy rate was 100%.
Abstract