Abstract:

We propose a votingbased multiclassifier method to improve the location accuracy of similar and distorted license plates. We enhance the license plate location accuracy from two aspects. First, two new plate descriptors are proposed according to the image characteristics of similar license plates and distorted license plates, which particularly promotes the location performance of the two types of license plates. Then several SVM classifiers are trained using different descriptors respectively, and a votingbased fusion mechanism is adopted to make the final decision, thus further enhancing the location accuracy.  Experimental results show that: 1) compared with the traditional wavelet and LBP license plate descriptors, the two proposed descriptors can effectively improve the location accuracy of distorted plates, and reduce the error rate of similar plates; 2) the votingbased fusion mechanism can effectively reduce the classification error rate of candidate plate regions from 3.05% to 0.8%.

Key words: classifier;descriptor;license plate location;Haarwavelet