模板匹配识别算法和神经网络识别算法的比较及MATLAB实现

Character Recognition Algorithm Based on Template Matching and Character Recognition Algorithm based on Neural Network Comparison and MATLAB Implementation

现有的车牌识别系统(LPR)已经诞生了许多关键技术。常用的关键技术有:基于数学形态学定位汽车车牌、基于Hough变换的车牌图像倾斜校正算法。而识别算法中,主要有模板匹配和BP神经网络算法。模板匹配算法是数字图像处理组成的重要部分之一。把不同的传感器在不同时间和成像条件下对景物获取到的图像在空间上对齐,或在模式到一幅图中寻找对应的处理方法。BP神经网络算法是一种“误差逆传播算法训练”:利用输出的误差估计前一层的误差,以此类推,获取各层次估计的误差。本次实验通过比较识别算法的两种关键技术得出以下结论:模板匹配实现过程简单,速度快,要求字符比较规整,并且对车牌图片质量要求很高,图像被其他因素干扰时,比如光线、清晰度等,会导致识别率低;而神经网络算法可以在不同的复杂环境下、不明确推理规则等识别问题,具有自适应性好、识别率高的自学习和自调整能力,但在识别前需要进行网络训练,速度慢,依赖大量的学习样本。

Existing license plate recognition(LPR)system has created the key technology of many key techniques are commonly used are:car license plate based on mathematical morphology on the license plate image tilt correction algorithm based on Hough transform and the recognition algorithm,there are mainly template matching template matching algorithm and the BP neural net⁃work algorithm is one of an important part of digital image processing of the different sensors in different time and imaging condi⁃tion of scenery get image alignment on the space,or in the model to a figure in search the corresponding treatment method the BP neural network training algorithm is a kind of error back propagation algorithm:using the output error of the estimation error of the previous layer,and so on,for all levels estimated error this experiment by comparing the recognition algorithm of two kinds of key technology in the following conclusions:template matching process is simple,fast,character is neat,and the license plate image quality requirement is high,the image interference by other factors,such as the light of clarity,leads to the recognition rate is low;However,neural network algorithm can identify problems such as unclear inference rules in different complex environments,and has self-learning and self-adjustment abilities with good adaptability and high recognition rate.However,network training is need⁃ed before recognition,which is slow and relies on a large number of learning samples.