首页> 外国专利> DESCREENING SYSTEM TRAINING METHOD AND APPARATUS, DESCREENING METHOD AND APPARATUS, DEVICE, AND MEDIUM

DESCREENING SYSTEM TRAINING METHOD AND APPARATUS, DESCREENING METHOD AND APPARATUS, DEVICE, AND MEDIUM

机译:侦听系统训练方法和装置,侦听方法和装置,装置和介质

摘要

Disclosed are a descreening system training method and apparatus, a descreening method and apparatus, a device, and a medium. The descreening system training method comprises: acquiring a first error function based on a screen pattern location and a pre-set tag value; acquiring a second error function based on a simulated picture and training samples without screen patterns; acquiring a third error function and a fourth error function according to a discrimination resu acquiring a fifth error function according to a feature A and a feature B; based on the characteristics of a generative adversarial network and according to the second error function, the third error function, the fourth error function and the fifth error function, respectively acquiring a sixth error function and a seventh error function; and then constructing an error function according to an error generated by training, and using the error function to carry out back propagation and update on network parameters of each model part in a descreening system. The descreening system removes multiple error factors affecting the descreening effect of the descreening system in a descreening system training process, so that the descreening effect of the descreening system is relatively good.
机译:公开了一种去网系统训练方法和装置,去网方法和装置,装置以及介质。所述拆网系统的训练方法包括:基于屏幕图案位置和预设标签值获取第一误差函数;基于模拟图片和训练样本而没有屏幕图案,获取第二误差函数;根据判别结果获取第三误差函数和第四误差函数;根据特征A和特征B获取第五误差函数;根据生成对抗网络的特征,根据第二误差函数,第三误差函数,第四误差函数和第五误差函数分别获得第六误差函数和第七误差函数;然后根据训练产生的误差构造误差函数,并利用该误差函数对去网系统中各个模型部件的网络参数进行反向传播和更新。该去网系统在去网系统训练过程中去除了影响去网系统去网效果的多种误差因素,因此去网系统的去网效果相对较好。

著录项

  • 公开/公告号WO2019200702A1

    专利类型

  • 公开/公告日2019-10-24

    原文格式PDF

  • 申请/专利权人 PING AN TECHNOLOGY(SHENZHEN) CO. LTD.;

    申请/专利号WO2018CN92569

  • 发明设计人 XU LINGLING;

    申请日2018-06-25

  • 分类号G06T5;

  • 国家 WO

  • 入库时间 2022-08-21 11:52:44

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号