ナンバープレート内の一連番号の切り出しと認識
- Author
- 佐藤省三, 藤吉弘亘, 梅崎太造, 金出武雄
- Publication
- 電気学会論文誌, vol. 121–C pp. 1354–1361, 2001
In order to segment robustly character areas of serial numbers in weak-contrast and noisy license plate images which are extracted from the front view of cars by the method described in our previous paper, we propose a useful method which is constructed by artificial neural networks corresponding to each digit of the serial numbers. The ability of proposed method is compared to that of a histogram-based method which is used as an example of conventional methods. In the histogram-based method, touching characters and noisy characters are often mis-segmented. The other hand, the network-based method can segment precisely even for those characters. For 595 license plate images, a segmentation rate of 96.7% in the histogram-based method and a rate of 100% in the network-based method with an additional second search are gained. Then, a network-based recognition method for the segmented characters is investigated. By randomly varying the segmentation size of trained characters in learning process of the network, all the serial numbers in the 595 images can be recognized.