摘要
首先,通过优化Cycle-GAN的结构与参数,提出改进Cycle-GAN算法以提高算法泛化能力;然后,提出基于归一化交叉相关(NCC)的训练数据集生长算法,使训练数据集能根据环境变化自动拓展;最后,在服务机器人上实现基于改进Cycle-GAN的服务机器人视觉图片隐私保护方法,并给出机器人视觉隐私图片保护系统.构建具有4.8×10~4张图片的对抗训练数据集,并建立6类测试实验的测试数据集对系统性能进行测试.实验结果表明:本文方法在转换成功率、体现图片相似度的哈希差值和体现脸部成像精度的人脸识别率方面均优于Cycle-GAN算法.
Firstly,focus on the generalization ability,an improved Cycle-GAN algorithm was proposed by optimizing the structure and parameters of Cycle-GAN.Then,a training dataset growth algorithm based on normalized cross correlation was proposed,which was used to automatically enrich the training dataset according to the environment change.Finally,the IPP-GAN was implemented on our developed social robot,and then the image privacy protection system of social robot vision was developed.In order to check the performance of the system,the source domain and target domain training dataset with 4.8×10~4 images were established,and the variate datasets based on 6 kinds of test solution were constructed.Experimental results show that the proposed method is superior to Cycle-GAN algorithm in success rate of image conversion,hash difference reflecting image similarity,and face recognition rate reflecting face accuracy.
作者
杨观赐
林家丞
李杨
李少波
YANG Guanci;LIN Jiacheng;LI Yang;LI Shaobo(Key Lab of Advanced Manufacturing Technology(Ministry of Education),Guizhou University,Guiyang 550025,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2020年第2期73-78,共6页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金项目资助项目(61863005,91746116)
贵州省科技计划资助项目(黔科合支撑[2019]2814,黔科合平台人才[2018]5702,黔科合平台人才[2018]5781)
贵州省2019年研究生科研基金资助项目(黔教合YJSCXJH[2019]109).
作者简介
杨观赐(1983-),男,教授,E-mail:gcyang@gzu.edu.cn.
