AdaBoost高速并行字符识别算法应用

Method of character recognition algorithm based on AdaBoost and its application

针对常用字符识别速度和精度矛盾的问题,提出了改进的AdaBoost字符识别算法。利用先验知识的稳定特征将字符集进行完全二分类,在此基础上分别训练级联的分类器,在充分的样本学习后可得到较高的识别正确率。针对AdaBoost算法的计算量大,用纯软件实现难以满足工业应用的实时性要求,根据其大量的乘累加运算相似性,基于积分图像和FPGA的并行结构来快速实现。实验结果表明,该算法能够满足印刷质量在线检测系统的识别正确率和实时性要求。

To solve the contradiction of speed and accuracy of commonly used methods of character recognition, an improved AdaBoost recognition algorithm is proposed. Based on full dichotomy of the character set by employing prior knowledge, cascaded classifiers are trained separately, generating high recognition accuracy by full sample learning. Because of high computation load, it＇ s hard to meet the real-time demand of industrial application by soft implementation of AdaBoost. Based on the similarity of mass multiply accumulation operation, parallel architecture based on FPGA is proposed. Application in online printing quality detection system demonstrated its reco- gnition accuracy and can meet the real-time requirements.