基于改进卷积神经网络的身份证信息识别

ID card information recognition with improved convolutional neural network

针对身份证信息识别中的文字定位和文字识别问题,提出基于改进卷积神经网络的身份证信息检测与识别方法。基于文字识别系统的主流框架并进行以下改进:针对目标检测网络定位文本精度低的问题,依据身份证的样式特点采用模板匹配方法进行文本定位。针对卷积核提取的特征表示能力弱的问题,采用残差模块结合卷积神经网络进行文字识别,使用预训练-微调范式解决证件数据集匮乏问题。针对文本间距离计算效率低的问题,采用集束搜索和语义处理优化识别结果。实验结果表明,该算法与对比方法相比,识别准确率有较大提升。

For the text localization and text recognition problems in identity(ID)card information recognition,an ID card information detection and recognition method based on improved convolutional neural network was proposed.Based on the mainstream framework of text recognition system,the following improvements were made.For the problem of low accuracy of text localization through target detection network,a template matching method was used to locate the text based on the style characteristics of ID card.To address the problem of weak feature representation ability extracted by convolutional kernel,the residual module combined with convolutional neural network was used for text recognition,and the pre-trained fine-tuning paradigm was used to solve the problem of lack of ID dataset.To address the problem of low efficiency of inter-textual distance calculation,beam search and semantic processing were used to optimize the recognition results.Experimental results show that the algorithm has improvement in recognition accuracy compared with the comparison methods.