智能系统学报2024，Vol.19Issue(6)：1479-1491,13.DOI:10.11992/tis.202307011
尺度可变有约束图像拼接检测与定位及其对抗优化
Scalable constrained image splicing detection and localization with adversarial optimizing
摘要
Abstract
A scalable detection and localization method,along with its adversarial optimization architecture,is proposed for the image forensics task of constrained image splicing detection and localization(CISDL).In the CISDL network,a novel scalable correlation computation module is constructed using high-efficiency channel attention enhancement blocks.The channel features are then calibrated using high-efficiency channel attention enhancement.Images of arbit-rary sizes are processed using truncation operations on closely correlated factors,and a mask reconstruction network is designed based on depthwise separable convolution and residual connections.Finally,a patch-level adversarial learning strategy is proposed to optimize the pretrained model.Extensive experiments on publicly available datasets demonstrate the effectiveness of the proposed method.关键词
有约束图像拼接检测与定位/尺度可变/关联性计算/对抗学习/图像取证/空洞卷积/金字塔池化/深度可分离卷积Key words
constrained image splicing detection and localization/scalable/correlation computation/adversarial learn-ing/image forensics/atrous convolution/pyramid pooling/depthwise separable convolution分类
信息技术与安全科学引用本文复制引用
刘亚奇,蔡强,石磊,张一凡,吕斌斌,夏超,许盛伟..尺度可变有约束图像拼接检测与定位及其对抗优化[J].智能系统学报,2024,19(6):1479-1491,13.基金项目
中央高校基本科研业务费资金项目(3282023016) （3282023016）
国家自然科学基金项目(62102010,62002003) （62102010,62002003）
北京工商大学食品安全大数据技术北京市重点实验室开放课题(BTBD-2022KF02). （BTBD-2022KF02）
