改进HOG特征的车牌识别算法

License Plate Recognition Algorithm Integrating Improved HOG Feature

为解决车牌图像中相似字符的误识别问题,提出一种融合改进局部HOG特征的模板匹配法。将改进方法与模板匹配法、模板匹配法结合跃变特征和模板匹配法结合局部HOG特征方法进行比较实验,测试第一类车牌图像时,识别率分别提高52%、12%和4%;测试第二类车牌图像时,识别率分别提高36%、28%和4%。改进的局部HOG特征确定了字符图像局部特征块的优化参数,得到局部优化的特征块,降低了特征描述符维数,在保证识别速率的同时提高了识别率。

In order to solve the problem of misrecognition of similar characters in license plate image,propose a template matching method integrating the improved local HOG feature to recognize license plate.Compared with the template matching method,the template matching method combined with jump feature and the template matching method combined with local HOG feature,the recognition rate of the method proposed increases by 52%,12%and 4%respectively when testing the first class license plate images.When the second type of license plate image is tested,the recognition rate is improved by 36%,28%and 4%respectively.The improved local HOG feature determines the optimal parameters of the local feature block of the character image,obtains the local optimal feature block,reduces the dimension of feature descriptor,and improves the recognition rate while ensuring the recognition rate.