A Simple Method for Chinese License Plate Recognition Based on Support Vector Machine

We present a simple method based on Support Vector Machine (SVM) for Chinese license plate recognition. By firstly pre-processing the input images containing license plates, a set of normalized subimages can be obtained, each of which contains a number, an English letter or a Chinese character. We then transform these subimages into vectors by simply using pixel values. In this way, we can avoid the problem of excessive dependency on feature extraction during recognition. Next, scaling and cross-validation are performed to eliminate outliers and find the best parameters for the SVM model. We use real color images captured at a motorway toll in our experiments. Be compared with previous work based on neural network, the SVM-based method produces a higher correct recognition rate. Experimental results also show the superiority of the SVM-based method when only a small number of samples are available.