摘要
为解决复杂自然场景下车牌字符受噪声等影响识别困难的问题,提出一种基于栈式降噪自编码神经网络的车牌识别方法。基于降噪自编码模型重构思想自动提取相关特征,通过使用无监督逐层贪婪预训练和有监督微调的方法对深度自编码神经网络进行训练,对复杂环境下低质量的车牌字符图像具有较好的鲁棒性能。与浅层的机器学习算法、传统栈式自编码神经网络和卷积神经网络相比,栈式降噪自编码神经网络有较好的字符识别性能。基于实际道口电子警察采集的车牌图像测试集的实验结果验证了该方法的有效性。
To solve problem of the license plate characters under complex natural scenes affected by noise and etc. , a method of the license plate characters recognition based on a stacked denoising autoencoder was proposed. Relevant features based on the reconstruction theory of denoising autoencoder were automatically extracted, and unsupervised greedy layer-wise pre-training and supervised fine-tuning were utilized to train the deep autoencoder network, so that it had good robust performance on obtaining the license plate characters with low quality in the complex environment. Compared with the shallow machine learning algo- rithms, the traditional stacked autoencoder and convolutional neural network, stacked denoising autoencoder is superior in recog- nition. Results of experiments of the license plate image test set collected by electronic police at the actual crossing verified the ef- fectiveness of the application method.
出处
《计算机工程与设计》
北大核心
2016年第3期751-756,共6页
Computer Engineering and Design
基金
江苏省自然科学基金项目(BK20130207)
江苏省博士后基金项目(1301029C)
关键词
车牌字符识别
栈式降噪自编码神经网络
重构
逐层贪婪预训练
微调
license plate character recognition
stacked denoising autoencoder
reconstruction
greedy layer-wise pre-training
fine-tuning
