Research On The Key Of The License Plate Recognition Based On AdaBoost Algorithm

It is well-known that it is very difficult to detect the low contrast license plate and it is impossible to detect multiply license plates for most of the automotive license plate recognition systems (LPRS) and the current LPRS is not good enough for the China character. This paper has present two contributions. One is the license plate detection system based on the Haar feature and the AdaBoost algorithm; another is the license plate recognition for China character based on the Gabor feature and AdaBoost_M2 algorithm.For the first part, this paper has gives detailed model design for the license plate detection system. And the experimental results has been shared, and these show that even the license plate contrast is very low, this system can detect the position of the license plate from the whole image, if you can choose a suitable scanning scale. And this system can detect multiply license plate from an image which including several license plates too. And in the end of this part, some other AdaBoost family algorithm such as DAB, RAB and GAB algorithm are used into this system for comparison. Some statistical experimental result has been present both in quantitative and qualitative analysis. And the comparison shows that GAB algorithm is better for license plate application.The second part is given for license plate recognition system about China character. Firstly, Through out China character typical feature, Gabor feature is chosen and some detailed design is given about Gabor filters. As we know there are 31 provinces in China and one character is stand for one province. So this is multiply class problem for this system. As a preferable good multiply classical algorithm, AdaBoost_M2 algorithm is chose for this task., and some statistical experimental results was given. And the experimental results shows that the recognition rate of this system has arrived to 91%, and basically reached the real-time level.